SEMANTIC NETWORK ANALYSIS IN

SOCIAL SCIENCES

Semantic Network Analysis in Social Sciences introduces the fundamentals of semantic

network analysis and its applications in the social sciences. Readers learn how to
easily transform any given text into a visual network of words co-occurring together,
a process that allows mapping the main themes appearing in the text and revealing its
main narratives and biases.
Semantic network analysis is particularly useful today with the increasing volumes of
text-based information available. It is one of the developing, cutting-edge methods to
organize, identify patterns and structures, and understand the meanings of our
information society. The first chapters in this book offer step-by-step guidelines
for conducting semantic network analysis, including choosing and preparing the
text, selecting desired words, constructing the networks, and interpreting their
meanings. Free software tools and code are also presented. The rest of the book
displays state-of-the-art studies from around the world that apply this method to
explore news, political speeches, social media content, and even to organize
interview transcripts and literature reviews.
Aimed at scholars with no previous knowledge in the field, this book can be
used as a main or a supplementary textbook for general courses on research
methods or network analysis courses, as well as a starting point to conduct your
own content analysis of large texts.

Elad Segev (PhD, Keele University) is Associate Professor at the Department of

Communication, Tel Aviv University. He studies the relationship between infor-
mation and power, focusing on global information flows, country image, interna-
tional news, information search, and the digital divide. In his studies he employs
text and network analysis techniques.
SEMANTIC NETWORK
ANALYSIS IN SOCIAL
SCIENCES

Edited by Elad Segev

First published 2022
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
and by Routledge
605 Third Avenue, New York, NY 10158
Routledge is an imprint of the Taylor & Francis Group, an informa business
© 2022 selection and editorial matter, Elad Segev; individual chapters, the
contributors
The right of Elad Segev to be identified as the author of the editorial material,
and of the authors for their individual chapters, has been asserted in accordance
with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or
utilised in any form or by any electronic, mechanical, or other means, now
known or hereafter invented, including photocopying and recording, or in any
information storage or retrieval system, without permission in writing from the
publishers.
Trademark notice: Product or corporate names may be trademarks or registered
trademarks, and are used only for identification and explanation without intent to
infringe.
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data
A catalog record has been requested for this book

ISBN: 978-0-367-63650-0 (hbk)

ISBN: 978-0-367-63652-4 (pbk)
ISBN: 978-1-003-12010-0 (ebk)
DOI: 10.4324/9781003120100

Typeset in Bembo
by Taylor & Francis Books
Access the Support Material: www.routledge.com/9780367636524
To my students
CONTENTS

List of figures ix
List of tables xi
Acknowledgments xii
List of contributors xiii

Introduction 1
Elad Segev
1 How to conduct semantic network analysis 16
Elad Segev
2 The news coverage of threats: Iranian nuclear programs in
Israeli press 32
Noa A. Hatzir, Elad Segev, Kohei Watanabe and Atsushi Tago
3 Provocation narratives in Chinese and US newspapers 53
Yuan Zhou, Sandrine Boudana and Elad Segev
4 Cable news channels’ partisan ideology and market share
growth as predictors of social distancing sentiment during the
COVID-19 pandemic 72
James A. Danowski, Bei Yan and Ken Riopelle
5 Politicizing the Holocaust: A comparative analysis of Israeli and
German speeches 94
Maya Hadar, Regula Miesch and Elad Segev
viii Contents

6 Network of cleavages?: 2British paradiplomacy in the (digital)

international discourse around Brexit 112
Jérôme Chariatte and Diana Ingenhoff
7 Sexual assaults blindsided by politics on Twitter: Semantic
Network analysis of symbolic #MeToo cases in Japan and
South Korea 136
Saki Mizoroki and Bumsoo Kim
8 Time to be happy: WhatsApp and phatic communication
within the extended family 159
Noa A. Hatzir
9 School improvement: Semantic network analysis of the
literature 175
Audrey Addi-Raccah, Tali Shahrabani and Elad Segev
10 Identifying patterns in communication science: Mapping
knowledge structures using semantic network analysis of
keywords 192
Nico Pfiffner
11 Summary and conclusion 216
Elad Segev

Glossary 228
Index 231
FIGURES

1.1 Frequent words from the first chapter of the Bible. 18

1.2 Converting frequent words of the first chapter in the Bible
into a link list. 19
1.3 Semantic network of frequent words in the first chapter of the
Bible. 21
1.4 The process of network sparsification. 26
1.5 Cluster analysis of the “fake news” discourse in elite
newspapers. 26
1.6 Cluster analysis of the #China discourse in Twitter. 29
2.1 Network of co-occurring actors in Haaretz news coverage on
Iran. 38
2.2 Network of co-occurring words in Yedioth Ahronoth reports on
Iran. 39
2.3 Network of co-occurring non-actor words in Haaretz reports
on Iran. 40
2.4 Network of co-occurring non-actor words in Yedioth Ahronoth
reports on Iran. 41
3.1 Monthly number of provocation articles in each newspaper. 59
3.2 The ratio of provocation related news per country in Chinese
and US media. 60
3.3 Undirected network of country co-occurrences in US news. 61
3.4 Undirected network of country co-occurrences in Chinese
news. 62
3.5 Semantic network of frequent words in US news. 63
3.6 Semantic network of frequent words in Chinese news. 64
x List of figures

4.1 “Partisan Press” mentions in Google books since the 1800s. 73

4.2 Social distancing frequencies and sentiment for Fox News,
MSNBC, and CNN. 81
4.3 Ideology, market share, and market share growth vs. social
distancing sentiment. 82
4.4 Social distancing semantic network CNN August 2020. 84
5.1 Semantic network of Netanyahu’s speeches mentioning the
Holocaust. 102
5.2 Semantic network of Merkel’s speeches mentioning the
Holocaust. 104
6.1 Semantic network of Tweets around #Brexit. 121
6.2 Semantic network of associations in the Irish media. 122
6.3 Semantic network of associations in the German media. 124
6.4 Semantic network of associations in the French media. 125
7.1 Network of co-occurring words in Japanese tweets for the
search term “Shiori Ito-”. 145
7.2 Network of co-occurring words in Japanese tweets for the
search term “Noriyuki Yamaguchi”. 147
7.3 Network of co-occurring words in tweets in Korean with
search term “Jihyeon Seo”. 148
7.4 Network of co-occurring words in tweets in Korean with
search term “Taeguen Ahn”. 150
8.1 Network of frequent words in two WhatsApp groups. 164
8.2 Network of frequent words in interview transcripts on
WhatsApp in the family. 167
9.1 Semantic network of words related to SI and SE. 180
9.2 Semantic network of words related to “teachers”. 182
10.1 Correlations between clustering coefficient and degree per
network. 203
10.2 Common structures across all networks. 206
TABLES

1.1 Free available tools for semantic network analysis 20

1.2 Identifying the most frequent and relevant words associated
with “fake news” 25
3.1 Top countries mentioned in provocation related news 60
4.1 Market share and growth for Fox News, MSNBC, and CNN 80
4.2 Higher August word pair frequencies for CNN 83
4.3 Higher August word pair frequencies for Fox News 83
4.4 CNN August word groups 86
4.5 Fox News August word groups 86
4.6 CNN August social distancing sentiment word strengths 87
6.1 Top 20 words with the highest degree centrality in all four
analyzed units 127
10.1 Communication journals in the sample 200
10.2 Structural network measures 202
10.3 Similarities of unweighted journal networks 205
10.4 Normalized weighted degrees of core nodes per network 208
10.5 Top keywords with the highest weighted degree centrality per
network 209
ACKNOWLEDGMENTS

This book is the result of teaching courses and workshops on network analysis for
more than a decade. It is therefore dedicated to my students in Israel, Switzerland,
England, and Japan. I am grateful to many of them who developed and shared
with me their passion for semantic network analysis. Endless thanks to Regula
Miesch, who, apart from her contribution to the research and writing one of the
chapters, also read and commented on the other book chapters and came up with
brilliant suggestions along the way.
Many thanks also to my dear colleague and contributors of this book: Noa
A. Hatzir, Kohei Watanabe, Atsushi Tago, Sandrine Boudana, Yuan Zhou,
James A. Danowski, Bei Yan, Ken Riopelle, Maya Hadar, Jérôme Chariatte,
Diana Ingenhoff, Saki Mizoroki, Bumsoo Kim, Audrey Addi-Raccah, Tali
Shahrabani, and Nico Pfiffner, many of whom worked with me on related
projects and agreed to share their studies in this book.
I am grateful to the anonymous reviewers who commented on the book proposal
and earlier drafts. Thank you for helping to make this book accessible and useful for
readers with no previous background in the field. Finally, I would like to thank the
editors at Routledge Research Methods, Hannah Shakespeare and Matthew
Bickerton, Jashnie Jabson and Martin Pettitt, for their dedication, patience, and
excellent guidance, making the publishing experience efficient and enjoyable.
CONTRIBUTORS

Elad Segev (PhD, Keele University) is an Associate Professor at the Department

of Communication, Tel Aviv University. He studies the relationship between
information and power from cross-national perspectives, focusing on global
information flows, country image, Americanization and globalization, international
news, information search and search strategies, and the digital divide. In his studies
he employs data mining and network analysis techniques.

Noa Ana Hatzir is a Junior Research Fellow and PhD student at the Department
of Communication, Tel Aviv University. Her research focuses on the intergenera-
tional communication, connections, and emotions in Israeli families, and how
media usage, which is increasing in our everyday lives, impacts intergenerational
communication. She also studied the reception of Israeli cinema in various cultures.

Kohei Watanabe (PhD, LSE) is an independent data analysis consultant for policy
think tanks and financial companies. He is also a visiting scholar at the London School
of Economics and Political Science (LSE) and Waseda University. He develops pop-
ular R packages for quantitative text analysis that include Quanteda, LSX and News-
map, and studies international flows of political and economic information.

Atsushi Tago (Ph.D., The University of Tokyo) is a Professor of International

Relations at the School of Political Science and Economics, Waseda University.
He is specialized in the scientific study of international politics. More specifically,
he publishes articles on American use of force and its coalition of the willing, mass
killing and public diplomacy. He is a PRIO Global Fellow.

Sandrine Boudana is a Senior Lecturer at the Department of Communication,

Tel Aviv University. Her research focuses on journalistic norms and practices, and
xiv List of contributors

news literacy, in a cross-national comparative perspective. Her recent publications

include “Comparative Studies”, The International Encyclopedia of Journalism
Studies (with A. A. Cohen) and “The language of objectivity: Reuters’ internal
editorial discussions on terminology”, Journalism (with G. Goodman).

Yuan Zhou is a PhD candidate at Kobe University. He specializes in international

relations and political communication, focusing on China. Methodologically, he
has employed a variety of quantitative methods in his research, especially quanti-
tative text analysis and Bayesian modeling.

James A. Danowski (Ph.D., Michigan State University) is an Emeritus Professor

of Communication at the University of Illinois at Chicago who theorizes about the
relationships between semantic networks and multi-level social networks in medi-
ated communication. He also develops new methods for communication research
using textual data, including the WORDij semantic network analysis package and
the SENET sentiment network analysis software.

Bei Yan is an Assistant Professor at the School of Business, Stevens Institute of Tech-
nology. She studies collaboration and influence processes of technology-supported
groups applying experiments, social network analysis, text mining. The core question
that underlies her research is how collective intelligence and structures emerge in
groups through computer-mediated communication and interaction with artificial
intelligence.

Ken Riopelle is an educator, entrepreneur, management consultant and retired

research professor at Wayne State University. His primary research interests include
accelerating the diffusion of innovations in globally networked organizations, the
study of collaborative innovation networks or COINs, and the science of team
science using co-author and co-citation analysis as a method to visualize, measure
and understand scientific collaboration.

Maya Hadar (PhD, LLM, MA Peace Studies and Conflict Resolution) is an

Assistant Professor of International Relations and European studies at Masaryk
University, the Czech Republic. A former visiting fellow at the Kilian Köhler
Center for Cultural Psychology and Historical Anthropology, Ruhr Universität
Bochum and a guest researcher at the Institute for Peace Support and Conflict
Management of the Austrian National Defense Academy. Her research interests
include nationalism and patriotism, identity politics, memory politics and mili-
tary-society relations in Israel.

Regula Miesch holds an MA in Translation Studies from Geneva University and

is a trained terminologist. She is fluent in German, English, French, Italian and
Hebrew. In a previous research project, she developed a multilingual dictionary to
study positive and negative views toward Israel in European newspapers.
 List of contributors xv

Jérôme Chariatte is a PhD student at the Department of Communication and

Media Research, University of Fribourg (CH). He is the holder of the “Master
Thesis Award for Excellence” of the European Public Relations and Research
Association (EUPRERA) and a “Top Faculty Best Paper Award” from the Public
Diplomacy Interest Group of the International Communication Association. His
research interests include international public relations, public diplomacy and
country images.

Diana Ingenhoff is a Full Professor of Organizational Communication and Public

Diplomacy at the University of Fribourg (CH). She is Executive Director at
EUPRERA and the founder and past chair of the first International Communica-
tion Association’s Public Diplomacy Interest Group. Her research and teaching
interests include recent developments in Public Diplomacy and International PR-
Research.

Saki Mizoroki is a PhD student at the University of Tokyo and a Visiting

Research Fellow at the Hebrew University of Jerusalem, Israel. Her research
focuses on feminist media studies, drawing on her experience as a veteran jour-
nalist. She has an MPP from the University of California, Berkeley.

Bumsoo Kim (PhD, The University of Alabama) is an assistant professor at

Joongbu University in South Korea. His research focuses on incivility in social
media, political polarization, and communication network heterogeneity.

Audrey Addi-Raccah (PhD, Tel Aviv University) is an Associate Professor at the

School of Education, Tel Aviv University. Her research is related to the sociology
of education and educational management with a focus on educational inequality,
teachers and school principals’ work and parental involvement in education. She is
currently engaged in large-scale research on school improvement and on social-
emotional learning in schools. In her research she uses mixed methodological
approaches, including large-scale data.

Tali Shahrabani is a PhD student at the School of Education, Tel Aviv Uni-
versity. Her dissertation focuses on Legitimate strategies of school principals. Since
2017 she is the coordinator of the unit of Sociology, Education and Community at
Tel Aviv University and teaches a course on social inequality.

Nico Pfiffner is a PhD student at the Department of Communication and Media

Research at the University of Zurich. His research interests revolve around the
factors influencing our digital media selection and, more generally, computational
social science.
INTRODUCTION
Elad Segev

We produce and consume a growing amount of information today on various media

channels. News is everywhere, people share their lives on social media, and bots are
programmed to promote political and economic interests. But how do we filter out
the “noise” and choose information that is relevant for us? How do we identify and
get rid of “fake news” and make sense of the world we live in? How do we reveal
the meaning of texts, their biases, and their underlying power structures?
These questions have always been important for our survival and in fact this is exactly
what our brain works on every second when we choose what to wear, what to eat,
how to act, and what to say or write to each other. With the availability of Big Data,
making sense of the information around us has become a real challenge. At the same
time, this is also a great opportunity for researchers who wish to understand this ever-
changing world to collect a tremendous amount of information and study its patterns.
The complex triangular relationship between politicians, news, and social media
is, for example, one of the trending research topics among political and commu-
nication scholars today. Different kinds of data, including the content of political
speeches, news content, and social media discourse, are constantly archived and
available for research. The relationship between commercial companies and their
brands, the advertisement industry, and the public views expressed on social media
is similarly an intriguing topic of study among economists and marketers.
In each of these examples, media and communication play a crucial role. As
scholars we are required to go through a very complex process in order to make
sense of this communication. We should identify first the most important actors
and information sources to be studied (e.g., relevant politicians, news outlets, and
influential social media accounts). This alone is a very challenging task. Once we
identify the desired information sources, we need to decide what information to
collect and how to analyze it. We are required to map the flow of information
(who said to whom?) and the content (what is being said?) of the interaction.
DOI: 10.4324/9781003120100-1
2 Elad Segev

Although pictures, images, and visual elements become increasingly important in

all these examples, in many channels text is still the predominant means of com-
munication, and words are still the prime building blocks. For a very long part of
human history, we have been combining words to create meanings and commu-
nicate our needs. A sentence such as “I am hungry” is a combination of words that
incorporates meaning. In practice, however, we have various (and sometimes
conflicting) needs on different levels. Thus, for example, the sentence “why is our
refrigerator always empty?” may be reduced to the same idea, but could as well
convey new social, economic, and even technological meanings. Studying the
choice of words, their unique combinations, and their contexts can therefore help
us to unveil hidden meanings and desires.
To address this challenge scholars have developed two main content analysis
approaches—quantitative and qualitative. While the quantitative approach focuses
on counting the reoccurring analytical units (e.g., constructing a codebook to
count how many times a text contains a reference to a certain topic and by whom),
the qualitative approach attempts to produce deeper layers of meanings and inter-
pretations to the text (e.g., using discourse analysis to study the explicit and implicit
references to power equalities). Yet both essentially aim at identifying patterns in
the text. Traditionally, these two approaches are treated separately: students take
separate courses in qualitative and quantitative research methods, and researchers
often employ only one of these two approaches in their studies. Yet qualitative and
quantitative approaches share many things in common, and this distinction may
limit the ability to exploit their full potential. In every quantitative process there
are various qualitative decisions that researchers are required to take. Thus, for
instance, there are qualitative reasons behind the construction of categories and the
assignment of words or sentences into one of them. Similarly, as qualitative
approaches aim at identifying patterns and reoccurring themes, research to some
extent “quantifies” and assesses whether patterns often appear or rarely in the text.
It would therefore be wrong to make a clear division between the two approaches
and treat them separately.
The matter gets more complicated when the available text gets larger. Manual
reading of text for qualitative and quantitative purposes is time-consuming. With a
lack of training on Big Data analysis, many scholars end up designing traditional
qualitative (such as discourse analysis) or quantitative (such as codebook) approaches
based on smaller samples of text when analyzing newspaper articles, political spee-
ches, or social media data. This is generally not a bad idea, as a proper understanding
of text always requires a close and careful manual reading. Yet, there are obvious
limitations, and scholars increasingly look for new and cutting-edge methods to
organize, identify the patterns and structures, and extract the meanings of this
increasing volume of information.
On top of this, the bigger picture is missing. With the abundance of information
available, it is often difficult for scholars to see how to navigate, where to start,
where to “dive-in”, and what to omit. A map or an overview of the content
highlighting the main actors, the prevailing and esoteric topics, and most
 Introduction 3

importantly how these topics relate or connect to each other would be most useful.
Eventually, scholars want to be able to say something meaningful about the infor-
mation they collect, to produce a meta-narrative that would support or challenge a
certain theory, or even generate a new explanation of the world we live in.
In this book, we aim to do exactly this. We focus on words as the building
blocks of content and communication processes and use a network approach to
map the relationships between words. The basic assumption behind this approach is
that our choices and combinations of words construct unique meanings. The
“actors” in the network are the words that frequently appear in the text. The “ties”
between words are their proximity or co-appearances, be it in the same sentence, a
post, tweet, or news item. Networks are therefore a way to translate a one-
dimensional linear text into a two-dimensional visual map. This approach, which is
often called textual or semantic network analysis, is useful in this sense as it enables
researchers to look at large texts from above, map the main themes, and understand
how they relate to each other before conducting more in-depth quantitative or
qualitative analysis. This book therefore focuses on the structure of the text, or
more precisely the network of words appearing together in the text.
The book was written for scholars and students with no previous background in
programming or network analysis. The first chapter provides the basic guidelines
and the publicly available tools to conduct semantic network analysis. The fol-
lowing chapters present the state-of-the-art research that applies semantic network
analysis in various fields. They are organized into three parts based on the type of
text we analyze. First, we demonstrate how semantic network analysis is used in
top-down communication, including the analysis of news media and political
speeches. Then we introduce semantic network analysis in bottom-up commu-
nication, including the analysis of social media and user-generated content. Finally,
we show how semantic network analysis can be used to process research data,
including meta-analysis of academic papers for literature review and the analysis of
interview transcripts to identify reoccurring themes for further qualitative analysis.

A short history of semantic network analysis

The relationship between words and their significance or meanings has been the
predominant subject of study in linguistics, literature, rhetoric, and more recently
in communication. Originating in Greek, the word “semantic” (σημαντικός)
refers to significance, and deals with the relationships between signs, symbols, or
words and their meaning. There have always been people who consciously looked
at the means of communication, systematically analyzed, classified, and carefully
studied their effectiveness. After all, effective communication provides better
chances for individuals and groups to promote their interests and goals, and more
generally to survive, reproduce, diffuse ideas, and exert control.
Yet, the combinations of words we express are also a reflection of the processes
in our minds. Our cognitive experiences, everything we see, hear, smell, taste, or
sense is coded and stored in our memory grid. The meaning of our experiences, as
4 Elad Segev

much as the meaning of words that signify them, is always processed in relation to
other experiences (Woelfel & Fink, 1980; Barnett, Palmer, & Noor Al-Deen,
1984). In other words, meaning is contextual; meaning is networks.
The idea that language is structured in a network of words that convey meaning
is not new. Throughout history, people have attempted to assess and classify the
functions of words, not only in terms of grammar, but also in terms of their
meanings, and social and political impact. In his Rhetoric, Aristotle describes three
modes of persuasion. Ethos refers to the speaker (being well-spoken and credible),
pathos refers to the hearer, the emotions, empathy, fear, and hope that the message
provokes, and logos refers to the arguments, reasons, and facts, and the ability of
the message to appear truthful. The three modes can overlap and affect each other.
The use of empathy (pathos) and scientific facts (logos), for example, can enhance
the credibility of the speaker (ethos).
Aristotle lived in a period and place where communication was largely oral and
face-to-face. However, his division seems to be very relevant today as well. With
the (growing) development of more mediated means of communication, these
three modes remind us that some of our communication is the attempt to persuade
others in the authenticity and truthfulness of the messages and their creators in
order to facilitate an action. When babies cry, they often need something, and their
parents are called to help. As we get older many of our needs become more
complex and so is our communication. Still, we all want that our needs will be
acknowledged and perceived as authentic or true. The triangular of ethos, pathos
and logos is therefore a very basic and universal way to understand the function of
words in the attempt to generate meaning, relevance, and impact.
Much more recently, the function of language attracted the attention of
anthropologists and linguists. When studying primitive languages, for instance,
Bronisław Malinowski (1923) noted that much of our communication is about
grooming, or what he termed as “phatic communication”. One of the frequent
functions of communication is not to exchange information but rather to signal the
willingness to be part of a community. Greetings, small talk, and body gestures aim
at simply establishing social networks. Still, if we wish to apply the rhetoric prin-
ciples, it is possible to argue that even in greetings and small talk people constantly
assess each other’s authentic intentions, and the wish to be loved and be a part of a
larger social network.
Perhaps one of the most prominent and influential theories of communicative
function is that of the linguist Roman Jacobson. Jakobson (1960) divided commu-
nication into six distinctive functions: referential, aesthetic/poetic, emotive, conative,
phatic, and metalingual. The referential function is related to the question of who
sends a message to whom. It is about the participants of the communication. The
aesthetic/poetic function deals with the form or style of the message. The emotive
function refers to the emotional aspect of communication, while the conative func-
tion refers to the imperative call for action. Finally, the phatic and the metalingual
functions refer to the ties, channels, and flows that connect people together. While
the phatic function focuses on establishing the channels of communication (through
 Introduction 5

greetings and small talk), the metalingual function is about the code, establishing a
common language of signs and meanings that are accepted by the participants.
Although Jacobson’s motivation was to describe the various dimensions of com-
munication processes rather than to highlight the art of rhetoric, there is a striking
overlap between the two classifications. Both referential and aesthetic functions focus
on the sender of the message (hence the ethos), their place in the communication
network (referential function), and their style (aesthetic/poetic). The emotive (emo-
tions) and conative (call for action) functions focus on the receiver (hence the pathos).
Although Jacobson’s classification does not emphasize the logo (the facts and reasons
embodied in the message itself), it adds a new dimension, which focuses on the
properties of the ties between people (the phatic and metalingual functions).
In parallel to the attempts to classify the functions of language and the meanings
of words, there were two crucial and inter-related developments in the field: the
use of computers in research and the emergence of network analysis. Since the
18th century, network analysis, also known as the “graph theory”, was mostly a
sub-field of mathematics with some minor applications in physics (Barabasi, 2002).
Yet, in the last few decades, network analysis has penetrated almost every possible
field of study, from biology to sociology, economy, linguistics, communication,
and political studies (Borgatti et al., 2009). The acceleration of globalization pro-
cesses and the increasing use of the internet in the 1990s have certainly contributed
to the realization that networks are an inevitable part of every aspect of our life.
The combination of content analysis and network analysis is relatively new, but
given the importance of both it seems to be indispensable. Scholars have gradually
realized the benefits of translating a linear text into a two-dimensional map, which
offers a fresh look at texts. Suddenly, it is not only the frequency of certain words,
but also the combination of words (also known as n-grams) that matters. But apart
from the relationships between words, networks allow us to scale up our under-
standing and look at the text as a holistic connected structure built from concepts,
ideas, and narratives. This is exactly what sematic network analysis is about.
At the beginning “sematic networks” was mainly a term used by language-
informed studies (Sagerer & Niemann, 2013; Sowa, 2014). During the 1970s and
1980s studies on semantic networks focused on the grammatical or ontological
functions of words and the logic behind their semantic relationships in specific
languages. Toward the end of the 20th century, however, the field of semantic
network analysis became a research paradigm of its own (Doerfel, 1998). It
encompassed not only the grammatical function of words but more broadly the
study of the relationships between words, concepts, and meanings.
In their review of the emerging studies in the field, Monge and Contractor (2001)
identified two different research traditions that develop from semantic networks
originating in two different research methods and approaches. One tradition focuses
on content analysis and therefore the actual words spoken or written and their rela-
tionships (Carley, 1993; Rice & Danowski, 1993). Another tradition is based on
surveys or in-depth interviews and therefore focuses on people’s associations and
interpretations (Monge & Eisenberg, 1987).
6 Elad Segev

In this book the focus is on content analysis. It is not about the interpretations of
ideas but more about the information produced and its social, cultural, and political
meanings. As will be demonstrated in the different chapters, semantic network
analysis is rooted in the social science and communication tradition of content
analysis, and therefore seeks to identify and analyze words within their theoretical,
social, and cultural (rather than grammatical or semantical) context. The combina-
tion of theoretically meaningful words that are frequently mentioned together leads
to the extraction of prevailing topics, frames, and biases in the text.

Between content analysis and network analysis

The approach of semantic network analysis as presented in this book lies at the
intersection between content analysis and network analysis. In content analysis the
aim is to classify the content into categories such as topics, sentiments, opinions,
and values. Some categories could be theory-driven and therefore a priori decided
in a top-down approach. Other categories could emerge from the bottom-up as
grounded theory. As will be demonstrated throughout the different chapters, net-
work analysis is on the meso-level in-between as it helps the researcher to better
understand the structure of the text and therefore to decide what top-down and
bottom-up categories to include.
The core idea behind network analysis is to focus on the interactions (sometimes
called “ties”, “links”, “connections”, or “edges”) between nodes (sometimes called
“actors” or “vertices”), and the structures that they entail. Thus, many studies
employing network analysis attempt to identify, quantify, and understand the
structure of the network, the impact of the strength of the ties between nodes
(Granovetter, 1977), the position of nodes within the network, and the patterns of
densely connected clusters (Borgatti et al., 2009). For example, the strength of the
tie can be measured by the amount and frequency of interactions (e.g., number of
conversations between two people), which can lead to trust between two people.
The centrality of a node can be measured by the number of connections it has
with other nodes (known as “degree centrality”), its distance from other nodes
(known as “closeness”), or its ability to mediate between other nodes (known as
“betweenness”). Different centrality measurements reflect different aspects of social
influence.
Clusters are often defined as relatively more dense sub-networks or commu-
nities. When we look, for example, at the network of social interactions in a city,
clusters would be smaller neighborhoods and communities as well as public and
private institutions, such as schools, hospitals, and firms. Identifying clusters helps
researchers to understand the cohesiveness in the network, which can reflect
social capital and trust, polarization, homogeneity, openness for new ideas, and
more generally the ability of the network to function as one unit. Similar to the
concept of network centrality, there are many different ways to identify com-
munities or clusters within a network, and this is currently an emerging field of
study (Hoffman et al., 2018).
 Introduction 7

While social network analysis can potentially unveil the power structures and the
way communities function, semantic network analysis is about revealing the structures,
the meanings, and the biases in the text people produce. In semantic networks the
nodes are the words or concepts, and the ties are their relationships. Central words
often appear in the text either with many other words (higher degree) or with other
specific central words (higher betweenness), and therefore they convey an important
meaning in the text. Similarly, when words are grouped into distinctive clusters, they
often share similar meanings, and thus they help researchers to classify the text into
smaller units of meanings and identify the prevailing themes and topics within the text.
Kathleen Carley and James A. Danowski were among the earlier scholars that
employed network analysis of words mentioned together in a text. During the
1990s, Carley (1993) introduced some fundamental guidelines for semantic net-
work analysis (termed “map analysis”), including the importance of choosing the
prominent words in the text, looking at the relationships between them, and
interpreting the meaning of these relationships based on previous social and cultural
knowledge. She also demonstrated the potential of semantic network analysis of
thesauri, fiction books, and even interview transcripts to highlight various aspects of
American culture (Carley, 1994).
Danowski (1993) further emphasized the main strength of semantic network
analysis compared to traditional content analysis methods. The use of codebooks,
for example, requires researchers to decide a priori what categories to look for in
the text. In semantic network analysis, categories emerge in a bottom-up process
based on the proximity between words (Rice & Danowski, 1993). Additionally, in
line with the “Aristotelian” approach, traditional qualitative and quantitative con-
tent analyses are based on a clear cut classification of content into exclusive cate-
gories. A network of words, on the other hand, would provide a continuous map
in which clusters of words overlap and relate to each other. In this way, researchers
can learn about the relative position of categories within the overall discourse. By
representing words as a continuum, semantic network analysis bridges between
qualitative and quantitative methods.
Semantic network analysis is less expensive and time-consuming than manual
coding with traditional codebooks or qualitative approaches. Researchers are not
required to manually code the text and conduct inter-coding reliability tests. As such
semantic network analysis fits well for Big Data studies and saves a lot of time and
resources. Moreover, even among the computerized methods of text analysis (see
below), semantic network analysis requires only a few resources and is considered
as one of the most straightforward and simple approaches. It is less complicated to
conduct than natural language processing, as it can be applied to any language,
without the need to “teach” the computer or “train” the algorithm with gram-
matical structures, or the semantic role of each word and part of speech.
Although semantic network analysis is a very useful tool to highlight the pre-
valent themes and map their relative positions and relationships, it can certainly not
replace the more in-depth qualitative analysis. Semantic network analysis is less
instrumental to understand the tone and valence, cynical expressions, hidden
8 Elad Segev

contexts, premises and assumptions, and underlying meanings. Similarly, as many of

the studies employing this method rely on frequent words or co-occurrences of
words, this method can be limited for extracting alternative and esoteric topics. It is
therefore recommended to use semantic network analysis along with other meth-
ods, and particularly qualitative content analysis, to get both a general and deeper
understanding of the text.

Semantic networks vs. other computational content analyses

In the recent years, the field of automatic content analysis has dramatically developed
together with the growing use of Big Data, artificial intelligence, and machine learn-
ing. Two major approaches are being used today to automatically analyze content:
supervised and unsupervised. In a supervised approach, researchers develop top-down
categories a priori, trying to study how often they appear in the corpus. They must
have some initial idea of what theories to use, which can be then operationalized in
measurable means. Thus, for example, when studying sentiments in the text,
researchers have to come up with a list of positive and negative “seed” words. In an
unsupervised approach, researchers do not always have a theory in mind. They try to
identify categories from the bottom-up based on the patterns that emerged from the
corpus. Semantic network analysis, as will be demonstrated later in the book, can be
one approach to find the themes appearing in the text without previously defining the
categories of analysis.
One of the easiest and most straightforward approaches for a supervised text
analysis is the “bag-of-words”. When using the “bag-of-words” approach
researchers should first define a list of words for each desired category. For exam-
ple, to determine if the text in question deals with sports, there is a need to define
first a list of words such as “ski”, “soccer”, “football”, and “match” that represent
the sports category. In this way the researchers can automatically assess whether a
document deals with sports and how often.
A more advanced method of text analysis is based on supervised machine learning.
The researchers read and manually classify documents, for example, emails as “spam”
or “not spam”. The computer uses this smaller sample of classified data (known as a
“training dataset”) to “learn” the differences between the documents in each cate-
gory. The learning process (known as the “model”) requires the selection of specific
parameters (also called “features”), such as the length of the articles, the existence of
unique frequent words in each category (e.g., “buy”, “earn”, or “cheap”), and the
email of the sender. Once the model distinguishes well between span and non-spam
emails in the training dataset, it could be applied to other much larger datasets.
An example of an unsupervised approach would be topic modeling. In this
approach there is no need to choose words or classify documents manually. The
computer first generates a list of frequent words for each document. Then it clas-
sifies the documents based on their word similarities. Documents dealing with
sports, for instance, are expected to share some unique keywords such as “football”
and “match” in common. The input in this case would be a large set of
 Introduction 9

documents, and the output would be groups of documents and a unique list of
words that are common to each group. The role of the researcher in such an
unsupervised procedure is to interpret the meaning of those groups of documents
based on the frequent words that are unique to each group.
Semantic network analysis is more of an unsupervised approach. It is similar to
topic modeling, but instead of classifying documents based on common words, it
classifies words within a document based on their proximity or co-occurrences.
The input in this case is a text, and the output is a network of words, clustered
based on their proximity. Semantic network analysis is considered to be an unsu-
pervised approach because there is no need to define in advance the categories, and
the classification of words into categories is done automatically based on their co-
occurrences within the text. Still, as will be explained in the next chapter,
researchers are required to have some idea of the theory or the research question in
mind as they choose relevant words to include in their networks.

From extraction to construction

Van Atteveldt (2008) described semantic network analysis as a process of extraction
and then construction of analytical units. In the extraction phase, researchers break
the text automatically or manually into small elements, which could be words,
phrases, or even larger concepts. In the construction phase researchers pay close
attention to the relationships between these elements as a whole network. During
this stage there is also a process of interpretation, in which the researchers attempt
to provide theoretical meanings to the clusters of words or concepts.
In fact, much of the scientific processes we know, whether empirical or theore-
tical, require breaking a larger phenomenon into smaller elements and then trying to
gradually find the relationships between those elements, and decomposing them
together in a certain way. During this process scientists gain better understanding of
the phenomenon in question, and sometimes manage to predict future cases, or even
manipulate and influence the course of events. Thus, for example, when marketers
wish to increase their online sales or journalists hope to attract more visitors to their
websites, they carefully study what were the ads or the news items, respectively, that
attracted the highest click traffic. They break those ads and news links into smaller
units, such as words, phrases and even signs (question or exclamation marks). From
that moment they start to look at the relationships between those elements, that is,
which combination of words work better. In other words, they look at the most
successful networks of words that bring more potential customers or readers. These
insights can help marketers and journalists to predict the success of future campaigns
or news items (or even design more effective communication).
Semantic network analysis helps us to automate this process and make it more
systematic. The smaller elements of the text are words, expressions, phrases, signs,
and even emojis. Their relationships are defined based on their physical proximity,
appearing two or three words away from each other, co-occurring in the same
sentence, the same post, the same paragraph, or the same document. Once the text
10 Elad Segev

is represented as a network, the structure of this network and its clusters can be
automatically obtained. Still, the last and most interesting and important stage of
the construction of meaning and assigning to each cluster theoretical frames is done
manually.

The application of semantic network analysis

Danowski (1993) pointed out the various applications of semantic network analysis,
starting from theory development, hypotheses testing, and identifying categories and
themes effectively. In terms of theory development, several studies informed by
grounded theory (Glaser & Strauss, 2017) used this method to map the relationships
and identify patterns within the information they collected from interviews, media
content, or even scientific publications. Thus, Bergstrom and Holmes (2000) con-
ducted interviews with bereaved spouses, and then inspected the word frequency and
the co-occurrences of words to understand the lay theory of successful aging after the
death of a spouse. Similarly, Veronese et al. (2018) employed semantic network ana-
lysis to identify reoccurring categories and themes in transcripts of in-depth interviews
related to teachers’ perceptions of wellbeing. Klenk and Larson (2013) combined
semantic network analysis with manual coding of hundreds of academic papers to
examine the co-occurrences of arguments around the debate over assisted colonization
of species among the scientific community.
Recent studies further employed semantic network analysis for hypothesis
testing and identifying categories and themes. Kang et al. (2017) examined the
co-occurrences of frequent words in Twitter to compare positive and negative
vaccine sentiment. Chung and Park (2010) compared the inaugural addresses of
two Korean presidents to highlight the different emphasis of conservative and
neo-liberal policies. Likewise, Meraz (2015) used semantic network analysis to
compare the different discourses and the existence of selective exposure among
progressive and conservative political blogs in the US during the 2012 presidential
election. Finally, Boudana and Segev (2017) explored the co-occurrences of country
names in major Western news outlets to identify the main countries involved in
provocation narratives. Their network clearly revealed the prominent international
conflicts reported by the Western media.
Paranyushkin (2018) further showed that networks of word co-occurrences in
the text can reveal not only different themes but also different levels of ideological
coherence based on their overall structure. Some semantic networks have dis-
tinctive clusters, while others are more centralized around few words. The cen-
tralized networks are typical for highly ideological and tendentious texts such as the
Quran that evolves around the word “god” or the Communist Manifesto that evolves
around the words “society” and “class”. Yet even in a highly centric network,
which focuses on one main issue, studies showed that the removal of one or two
dominant words could often help to reveal a more distributed structure of word
clusters, representing different themes or frames. Thus, when analyzing the co-
occurrences of country names in international news, Segev and Blondheim (2013)
 Introduction 11

found that the US is by far the most central. Only by removing the US from the
analysis was it possible to detect clear regional clusters of international interactions.
To conclude, the applications of semantic network analysis are numerous. Some
previous studies have applied semantic network analysis to develop theory (Berg-
strom and Holmes, 2000; Klenk and Larson, 2013; Veronese et al., 2018), others
examine a hypothesis (Chung and Park, 2010; Kang et al., 2017; Meraz, 2015;
Segev and Blondheim, 2013), and conduct initial mapping of the main themes and
topics in the text (Carley, 1994; Boudana and Segev, 2017). In this introduction,
the strengths of semantic network analysis were presented, including the ability to
perform a more objective grounded content analysis, in which the text is trans-
lated into a two-dimensional map. Moreover, unlike traditional qualitative and
quantitative content analyses, the emerging themes and topics are positioned in a
continuum with a clear relationship to each other, rather than as discrete units. In
this way, semantic network analysis provides a bridge between quantitative and
qualitative methods (Danowski, 1993).

The structure of this book

This book attempts to address the growing need for new methods to deal with the
vast amount of information available for social scientists. Chapter 1 introduces the
basic guidelines for conducting semantic network analysis. It provides step-by-step
instructions for students and researchers who have no previous knowledge of the
subject, presents some of the publicly available tools, and demonstrates their use
when conducting such an analysis. The rest of the chapters in the book present
various applications of semantic network analysis in different research fields, divided
into three parts: semantic network analysis in top-down communication (including
traditional news outlets and political speeches), semantic network analysis in
bottom-up communication (including social media and user-generated content),
and semantic network analysis in research data (including meta-analysis of academic
papers and the analysis of interview transcripts).
In Chapter 2, Noa Hatzir, Kohei Watanabe, Atsushi Tago, and I (Elad Segev)
demonstrate how semantic network analysis can be used to unveil news narratives in
Israel related to the Iranian nuclear threat. For many decades nuclear threats have
been at the top of the international agenda. In this chapter we analyze the content of
two prominent Israeli newspapers with different political inclinations when reporting
on the development of Iran’s nuclear program over a ten-year period. Using
semantic network analysis, we identify the main topics and issues that are raised in
the context of the Iranian nuclear plan, and the role of certain media outlets to
promote the national political agendas of the Israeli government in power.
In Chapter 3, Yuan Zhou, Sandrine Boudana, and I display how semantic network
analysis can be used to study the more specific discourse around certain keywords. We
chose the word “provocation”, which is often used to frame troublemakers in inter-
national news and justify reaction. We show that provocation narratives inevitably
introduce political bias. Semantic network analysis helps us to compare between the
12 Elad Segev

provocation narratives in American and Chinese newspapers. First, we identify the

core countries mentioned in the context of provocation. Then we look at the
prevailing themes and topics in American and Chinese news. Our findings show
that provocation narratives are typically used to highlight the major international
conflicts around the world, and particularly the conflict between North Korea,
South Korea, and the US. While American news emphasizes the role of North
Korea as the provocateur, Chinese news emphasizes the role and responsibility of
the US. This indicates that by using provocation narratives, local media adopt the
official government agendas, attempting to portray their enemies or competitors as
a global threat.
In Chapter 4, James A. Danowski, Bei Yang, and Ken Riopelle apply semantic
network analysis to understand the news discourse around the term “social distan-
cing”, which became very prominent during 2020. Unlike the previous two
chapters, which are based on print news, in this chapter the corpus is based on
transcripts of three television channels, Fox News, MSNBC, and CNN. As these
channels differ in terms of political orientation and market share, they expected to
find a different news coverage on this topic. Another methodological innovation is
that semantic network analysis was employed to determine more accurately the
positive and negative sentiments in the coverage of social distancing. They found
that the market share growth of the different TV channels was a better predictor of
the sentiment level than their political orientation. In the Appendix they also detail
the process and provide the code in R for the data preparation, analysis, and
visualization.
In Chapter 5, Maya Hadar, Regula Miesch, and I show how semantic network
analysis can be useful for highlighting the biases in political speeches. We examine and
compare public speeches made by the Israeli Prime Minister Benjamin Netanyahu and
the German Chancellor Angela Merkel mentioning the Holocaust over the last
decade. Semantic network analysis enabled us to identify the main themes raised by
each politician and how these themes help each one of them to promote their main
narrative on the identity and current role of Israel and Germany, respectively. The
findings in this chapter contribute to the growing literature on the politics of memory
and the way the political elite utilizes collective memories and narratives to promote
political goals and shape national identity.
The second part of the book is devoted to studies that make use of semantic
network analysis of social media content. In Chapter 6, Jérôme Chariatte and Diana
Ingenhoff apply semantic network analysis in the field of digital city diplomacy and
country image formation. They investigate the discourse on Brexit, the withdrawal
of the UK from the European Union, which not only challenges its international
relationships, but also divides the British nation. While populist voices support the
referendum, cities like London or whole nations like Scotland communicate their
affection for Europe. Social media in particular makes it possible for sub-actors to
participate in diplomatic discourse. But are these voices actually heard in the inter-
national arena? And to what extent can a country position itself independently in
times of global dependencies? This chapter compares the discourse in international
 Introduction 13

news with the discourse on Twitter. It uses semantic network analysis to extract
central thematic associations and patterns from large data sets. Their findings
show that paradiplomacy does indeed shape public discourse and that non-state
actors, cities, and regions can significantly contribute to the standing of a country
and its issues.
In Chapter 7, Saki Mizoroki and Bumsoo Kim apply semantic network analysis to
study a large corpus of tweets dealing with two widely publicized rape cases in Japan
and South Korea. While gender-based harassment and feminist hashtag activism have
received scholarly attention, it is relatively limited to Anglophone users or single-
country studies. To compare the tweets in Japanese and Korean, they retroactively
collected data between 2017 and 2020, using the names of female victims and male
assailants. This chapter contributes to the literature on social media and gender,
networked misogyny, fragile masculinity, and personalization of politics.
Finally, the studies presented in the third part of the book focus on the appli-
cations of semantic network analysis in research data. In Chapter 8, Noa Hatzir
studies the perceived and actual role of WhatsApp in the extended family. She
conducts in-depth interviews, analyzes the content of two WhatsApp groups, and
employs semantic network analysis to identify the main themes not only in social
media content but also in interview transcripts. Her findings reveal that within the
extended family in Israel WhatsApp is perceived as a useful and often preferable
channel. This is mainly due to the variety of communication forms that it offers
and the ability of users to control the timing of conversations. It is therefore used
also for checking people’s availability as well as organizing meetings. The actual
WhatsApp conversations among members of the extended family focus primarily
on positive emotions, with an emphasis on celebrating birthdays, holidays, sche-
duling family gatherings, and sharing photos. Hence, it appears that the use of
WhatsApp encourages mostly image management and phatic communication.
In Chapter 9, together with Audrey Addi-Raccah and Tali Shahrabani, we apply
semantic network analysis to review and map the recent academic literature on school
improvement. School improvement research is highly prolific and diverse in the
attempt to understand the processes that lead to effective and successful schooling. We
analyzed more than 200 academic papers published in the last five years dealing with
school improvement, successful, and effective schooling. We then employed semantic
network analysis to identify reoccurring themes and topics. Our findings show that the
main narrative of school improvement research is a conservative one, in which the
responsibility lies in the hands of the teachers and managers, while the students, parents
and other external stakeholders are viewed as more passive in the process. The edu-
cational goals stress improvement over effectiveness. School improvement has been
primarily defined and assessed based on students’ academic abilities and their outcomes
or scores in (particularly math) tests. There is much less emphasis in recent literature on
life skills, social capital, or social and universal values and school collaboration with
external agencies.
In Chapter 10, Nico A. Pfiffner uses a different approach to map the field of
communication research by looking at the keywords in academic papers. To do so,
14 Elad Segev

he extracted all papers published in the last 15 years in leading communication

journals, collected author-assigned keywords, and constructed a network of key-
words based on their co-occurrences in the same paper. This network serves as a
representation of the knowledge structure of the field of communication science.
Based on the structural characteristics of the keyword network, prominent topics in
the field are identified and the relations between different topics are investigated.
Furthermore, the developments and changes of the thematic focus over the course
of the observed period are shown and discussed.
While these chapters vary in terms of discipline, scope, theories, types of datasets,
and contexts, they all have one thing in common: the application of semantic net-
work analysis. They explore the associations between chosen keywords to identify
patterns, prevalent themes, and underlying meanings within the text. Since news data
are very different from social media data or interview transcripts, at the end of each
chapter there is a special section dedicated to tips and lessons that can be drawn from
the process. We hope that by showing the advantages and limitations of semantic
network analysis in different settings we will encourage researchers from different
backgrounds to try, apply this approach, and develop it further.

References
Barabasi, A. L. (2002). Linked: The New Science of Networks. Cambridge: Perseus.
Barnett, G. A., Palmer, M. T., & Noor Al-Deen, H. (1984). An examination of the use of
multidimensional scaling to determine the accuracy of translation. In R. Bostrom (Ed.),
Communication Yearbook (vol. 8, pp. 659–679). London: Routledge.
Bergstrom, M. J., & Holmes, M. E. (2000). Lay theories of successful aging after the death of
a spouse: A network text analysis of bereavement advice. Health Communication, 12(4),
377–406.
Boudana, S., & Segev, E. (2017). The bias of provocation narratives in international news.
The International Journal of Press/Politics, 22(3), 314–332.
Borgatti, S. P., Mehra, A., Brass, D. J., & Labianca, G. (2009). Network analysis in the social
sciences. Science, 323(5916), 892–895.
Carley, K. (1993). Coding choices for textual analysis: A comparison of content analysis and
map analysis. Sociological Methodology, 23, 75–126.
Carley, K. (1994). Extracting culture through textual analysis. Poetics, 22(4), 291–312.
Chung, C. J., & Park, H. W. (2010). Textual analysis of a political message: The inaugural
addresses of two Korean presidents. Social Science Information, 49(2), 215–239.
Danowski, J. A. (1993). Network analysis of message content. In W. D. Richards & G. A.
Barnett (eds.), Progress in Communication Sciences, Vol. 12 (pp. 198–221). Norwood: Ablex
Publishing Corp.
Doerfel, M. L. (1998). What constitutes semantic network analysis? A comparison of
research and methodologies. Connections, 21(2), 16–26.
Glaser, B. G., & Strauss, A. L. (2017). Discovery of Grounded Theory: Strategies for Qualitative
Research. London: Routledge.
Granovetter, M. S. (1977). The strength of weak ties. In S. Leinhardt (Ed.), Social Networks
(pp. 347–367). London: Academic Press.
Hoffman, M., Steinley, D., Gates, K. M., Prinstein, M. J., & Brusco, M. J. (2018). Detecting
clusters/communities in social networks. Multivariate Behavioral Research, 53(1), 57–73.
 Introduction 15

Jakobson, R. (1960). Linguistics and poetics. In T. Sebeok (Ed.), Style in Language (pp. 350–377).
Cambridge, MA: MIT Press.
Kang, G. J., Ewing-Nelson, S. R., Mackey, L., Schlitt, J. T., Marathe, A., Abbas, K. M., &
Swarup, S. (2017). Semantic Network Analysis of Vaccine Sentiment in Online Social Media.
Vaccine, 35(29), 3621–3638.
Klenk, N. L., & Larson, B. M. (2013). A rhetorical analysis of the scientific debate over
assisted colonization. Environmental Science & Policy, 33, 9–18.
Malinowski, B. (1923). The Problem of Meaning in Primitive Languages. In C. K. Ogden
and I. A. Richards (eds.), The Meaning of Meaning (pp. 296–336), London: Kegan Paul,
Trench and Trubner.
Meraz, S. (2015). Quantifying partisan selective exposure through network text analysis of
elite political blog networks during the US 2012 presidential election. Journal of Informa-
tion Technology & Politics, 12(1), 37–53.
Monge, P. R., & Contractor, N. S. (2001). Emergence of communication networks. In F.
Jablin & L. L. Putnam (Eds.), The New Handbook of Organizational Communication: Advances
in Theory, Research, and Methods (pp. 440–502). Thousand Oaks, CA: SAGE Publications.
Monge, P. R., & Eisenberg, E. M. (1987). Emergent communication networks. In F. Jablin,
L. L. Putnam, K. Roberts, & L. Porter (Eds.), Handbook of Organizational Communication
(pp. 304–342). Newbury Park, CA: SAGE Publications.
Paranyushkin, D. (2018). Measuring discourse bias using text network analysis. Towards Data
Science. https://towardsdatascience.com/measuring-discourse-bias-using-text-network-ana
lysis-9f251be5f6f3.
Rice, R. E., & Danowski, J. A. (1993). Is it really just like a fancy answering machine?
Comparing semantic networks of different types of voicemail users. Journal of Business
Communication, 30, 369–397.
Sagerer, G., & Niemann, H. (2013). Semantic Networks for Understanding Scenes. Erlangen:
Springer Science & Business Media.
Segev, E., & Blondheim, M. (2013). America’s global standing according to popular news
sites from around the world. Political Communication, 30(1), 139–161.
Sowa, J. F. (Ed.). (2014). Principles of Semantic Networks: Explorations in the Representation of
Knowledge. Burlington, MA: Morgan Kaufmann.
Van Atteveldt, W. H. (2008). Semantic Network Analysis: Techniques for Extracting, Represent-
ing, and Querying Media Content. Charleston, SC: BookSurge.
Veronese, G., Pepe, A., Dagdukee, J., & Yaghi, S. (2018). Teaching in conflict settings:
Dimensions of subjective wellbeing in Arab teachers living in Israel and Palestine. International
Journal of Educational Development, 61, 16–26.
Woelfel, J., & Fink, E. L. (1980). The Measurement of Communication Processes: GALILEO
Theory and Method. New York: Academic Press.
1
HOW TO CONDUCT SEMANTIC
NETWORK ANALYSIS
Elad Segev

The process of semantic network analysis generally consists of two main stages: text
preparation and analysis. First, researchers are required to choose the right text and
prepare it for analysis. Some texts, such as instant messages in a chat group, may be
less informative, and the content of this text would therefore be of no interest.
Other texts may not fit the research questions. The corpus should be well defined
and properly prepared for analysis, including the selection of desired words to be part
of the network analysis. In the second stage, the network is constructed, analyzed,
and interpreted. This stage includes simplifying the network structure and identifying
the clusters and the most central words in the network, which represent the main
discourse and narratives appearing in the text. The following steps are recommended
to be taken as part of the semantic analysis:

Stage 1 Preparing your corpus for analysis

1.1 Defining and refining the sample

The first stage requires a clear definition of the corpus to be studied and its rele-
vance to the research question. When researchers have a clear question in mind, it
is sometimes sufficient to skim through the text to see if the answer might be
hiding in it. When research questions are not yet formulated, skimming through
the text enables researchers to assess whether there are interesting issues that could
be raised and highlighted. As a rule of thumb, a good text is a provocative text. It is
a text that contains debates, tensions, contradictions, biases, explicit, or implicit
agendas. It should be rich and diverse enough in terms of vocabulary and the span
of topics to allow a meaningful analysis. Thus, for example, literature and art, news
articles, political speeches, social media debates on Twitter or Facebook, Wikipedia
articles, transcripts of interviews or films, religious scripts, and even academic papers
DOI: 10.4324/9781003120100-2
 How to conduct semantic network analysis 17

may be appropriate for semantic network analysis. The next chapters in the book
will demonstrate how to analyze different types of texts and the strengths and
limitations of such an analysis.
In the last two decades various new sources for text analysis have become
available to researchers, such as Factiva and Nexis Uni databases for news, legal,
government, and business information, social media content on Twitter and Face-
book API, blogs and Wikis, search engine results, and so forth. Together with new
sources, social scientists rely also on traditional sources such as transcripts of in-
depth interviews and collections of research articles. As the amount of data
increases, semantic network analysis offers an innovative and fresh approach for
both new and more traditional types of data, as it allows researchers to organize,
visualize, and obtain an initial map of the main topics and themes appearing in
those corpora.

1.2 Selecting the words to be analyzed

The second stage is crucial as it dramatically affects the rest of the analysis. In
this stage researchers need to choose the keywords, expressions, or phrases to
be studied in light of the theory and research questions. In traditional discourse
analysis researchers closely read the text to identify those key terms. Reading
through the text is still a very important stage in semantic network analysis, as
it helps knowing better the topics, use of words, and the communication
dynamic. It also helps in revealing central words that appear in specific contexts
and may have essential cultural and social meanings. Thus, for example, when
reading through political speeches, one may find the word “I” to be very
important, as it reflects the rhetoric and position of the speaker (see Chapter 5).
In many other types of text the word “I” is merely an unnecessary stop-word
that could be ignored.
Together with these qualitative methods, a common computer-based
approach would be to construct a list of frequent words in the text and choose
the top 100–200 words. This can be done with the help of freely available
online tools (see Table 1.1). Another option would be to focus on specific
words such as names of people, companies, organizations, or countries men-
tioned in the text. This could help to reveal the dynamic between the actors
involved in the discourse. In this stage researchers often combine top-down and
bottom-up approaches. From the top-down perspective the theory, research
questions, and hypotheses should always be in mind when choosing the rele-
vant keywords. From the bottom-up perspective, researchers should be open to
including keywords that add new categories and fresh dimensions to the
analysis.
Figure 1.1 demonstrates how to extract the most frequent words in the first
chapter of the Bible, using the word frequency calculator available online from
Segev’s (2020) set of text analysis tools.
18 E. Segev

FIGURE 1.1 Frequent words from the first chapter of the Bible.

It is important to note, however, that before constructing a list of the most fre-
quent words the text might need to undergo some initial processing and prepara-
tion. In most languages, words are separated by spaces, which enables algorithms to
automatically detect and count them as separate units. In some languages, such as
Chinese or Japanese, researchers should first use software to separate words from
each other (a process known as “tokenization”, see also the examples in Chapters 3
and 7).

1.3 Cleaning and refining the list of words

Apart from commonly used words (pronouns, prepositions, and conjunctions such
as “the”, “that”, or “of”, which are known as “stop-words”), there are often fre-
quent words in the text (such as “one” or “able”) that may appear in different
contexts and have little or no relevance to the theory and research question. When
constructing networks of words co-occurring in the text, these general words often
appear at the center, and thus prevent from identifying the overall structure of the
network and the existence of clusters of words. They should be therefore removed
from the final list before proceeding to the next stage.
Another important action to be taken at this stage would be merging families of
derivationally related words with similar meanings (also known as “stemming” and
“lemmatization”). For example, conjugations (e.g., “believe” and “believing”),
single and plural forms (e.g., “president” and “presidents”), or prepositions that, in
 How to conduct semantic network analysis 19

some languages, can be part of the same word, should be merged. Similarly,
common expressions such as “social media” or “united states” should be combined.
One can decide to create several lists, each with a focus on different aspects of the
text. Thus, for example, selecting nouns and expressions could be useful when
attempting to identify the most prevailing topics (Rule et al., 2015). The focus on
adjectives could be particularly useful when the research aim is to study sentiments,
opinions, and views, while verbs are useful to highlight actions.

Stage 2 Constructing and analyzing the network

Converting the text and list of words to a network. The unique power of semantic
network analysis is its focus on the interactions between words and their co-
occurrences to produce meaning. Once the clean list of about 100–200 words
has been finalized, it is possible to generate a link list or a matrix of word pairs
that co-occur together. For this purpose, one should first define the “window”
(such as a sentence, a post, a tweet, a paragraph, or even a news item), in
which each pair of words should be counted. Some free online tools (see Table
1.1) enable converting any text to a link list format given the desired list of
predefined words. While words act as the nodes in the network, the ties
between them are the number of sentences, posts, paragraphs, or news items in
which they appear together.
Figure 1.2 demonstrates how to use Segev’s (2020) set of text analysis tools to
convert a list of desired frequent words, appearing at least twice in the first chapter
of the Bible, into a link list. This link list contains word pairs and their frequency of
co-occurring in the same sentence.

FIGURE 1.2 Converting frequent words of the first chapter in the Bible into a link list.
20 E. Segev

Another way to link words together would be based on their distance or proxi-
mity, a method also known as k-next-neighborhood (see, for example, Drieger,
2013; Paranyushkin, 2011). Danowski (1993) recommended the construction of
word pairs with a distance of three words in between for optimal identification of
clusters. Finally, another approach suggested by van Ham, Wattenberg, and Viégas
(2009) is to focus on specific grammatical patterns to explore in detail the various
dimensions of the text. They show, for example, that when scanning the Bible for
textual matches to the pattern “X begat Y”, it is possible to identify the network of
family relations, while a pattern such as “X’s Y” (e.g., God’s children) can be used to
reveal power relations.

2.1 Network preparation

The link list or matrix obtained could be saved as a CSV file (comma-separated value)
directly from Segev’s (2020) set of text analysis tools. Alternatively, one can use Excel or
any other spreadsheet software to save the link list of word pairs as a CSV file and then
import and open it from the most freely available open-source network software tools
(see Table 1.1). Network software tools such as Gephi, NodeXL, Visone, or KH
Coder, as well as R or Python packages, can read a CSV file containing a link list or a

TABLE 1.1 Free available tools for semantic network analysis

Stages Web-based and software R and Python packages

1.1 Define and refine the
sample
1.2 Identify the main actors
in the network
1.3 Clean and refine the list
2.1 Convert the text and list
of words to a network
2.2 Network sparsification
2.3 Identify clusters and the
structure of the network
2.4 Identify the central words
(require programming
skills)
News: Factiva, Nexis Uni R packages for most
Social Media: DMI, Mozdeh, TAGS stages: quanteda and
For identifying frequent words: textnets
Semantic network analysis, Mozdeh, Python packages:
Count Words, KH Coder Text-Network Ana-
lysis or a combination
Use semantic network analysis to of the nltk and
clean stop-words and then manually NetworkX
choose relevant words
Semantic network analysis, Visone
with NLP extension, KH Coder
Visone, Gephi, NodeXL

Sources: General semantic network analysis tools: http://www.eladsegev.com/tools, Count Words: http://
www.countwordsfree.com, Digital Methods Initiative (DMI): https://wiki.digitalmethods.net/Dmi/
ToolDatabase, Mozdeh: http://mozdeh.wlv.ac.uk (Thelwall, 2018), TAGS: https://tags.hawksey.info,
Visone: https://visone.info, Gephi: https://gephi.org, NodeXL: https://nodexl.com, quanteda: https://qua
nteda.io/reference/textplot_network.html (Benoit, 2018), textnets: https://github.com/cbail/textnets (Bail,
2016), Text-Network Analysis: https://github.com/michal-pikusa/text-network-analysis, KH Coder: https://
khcoder.net/en/ (Higuchi, 2001).
 How to conduct semantic network analysis 21

matrix of word pairs and translate it into a network of words. Basically, these tools
enable us to visually turn the linear one-dimensional text into a two-dimensional net-
work and make all the necessary calculations (as described below) to identify the most
central words in the network and the clusters of words frequently appearing together.

2.2 Network sparsification

The network graph that appears initially is often very dense, messy, and therefore
unreadable and difficult to interpret. There are several ways to sparsify the network so
that it will be more “readable” or meaningful for the researcher. One would be to
delete the most central word in case it is linked to most other words. This typically
occurs when the text is obtained based on a specific search term. Thus, for instance,
when looking for all the tweets or posts that mention the word “happiness”, the word
“happiness” will appear at the absolute center of the semantic network, connecting to all
other words. Deleting this word will often break the network into clusters and enable
researchers to detect the main topics discussed in social media in relation to happiness.
Another way to clean the network would be to merge or unify similar words
(such as singular and plural forms, alternative spellings, or different names of a
similar entity). Finally, if the network is still too dense and it is difficult to detect its
structure one could remove less frequent links. As a rule of thumb it is easier to
analyze a network of 100 words when it contains between 200 and 400 most fre-
quent links. Lindner et al. (2015) provide a further review and analysis of more
sophisticated methods and algorithms to sparsify the network, while preserving its
overall structure.

FIGURE 1.3 Semantic network of frequent words in the first chapter of the Bible.
22 E. Segev

Using Segev’s (2020) set of text analysis tools, Figure 1.3 shows the semantic
network of frequent words in the first chapter of the Bible based on the 50 most
frequent word pairs. This visualization already enables us to identify two main
words: “god” and “earth” as the most central words in the first chapter. The word
“earth” is linked to many words coming from the agriculture tradition such as
“seed”, “tree”, “fruit”, “water”, “beast”, and “cattle”. Interestingly, however, the
word “god” mediates between the lower agricultural world and the upper uni-
verse, which provides the conditions for life and growth, that is “light” vs. “dark-
ness”, “day” vs. “night”, and “morning” vs. “evening”.

2.3 Identify clusters (themes) and the structure of the network

Once the network’s structure appears and clusters are visually identifiable, many
network software tools offer the possibility to perform a cluster analysis to
identify the main themes and frames that automatically emerge based on word
co-occurrences. Computer and information science researchers currently invest
much effort in exploring innovative ways to identify meaningful clusters of
words. One of the recently developed optimal methods for cluster analysis
within the network is Louvain modularity, as it measures the density of links
within communities (Blondel et al., 2008). Most network software implements
the Louvain modularity. For a network of about 100 words with 100–200
links, this method often produces a manageable number of clusters, reflecting
the inner network structure and the most prevailing themes and issues appear-
ing in the text.
Apart from revealing the main themes and their relative position Danowski
(1993) suggests to inspect the most prevalent paths between words, which can
unveil the main narratives appearing in the text. When looking at the prevailing
paths between words in Figure 1.3, for example, one can easily observe at least
three parallel main narratives in the first chapter of the Bible. The first path
represents the act of creation itself with words such as “god”, “said”, “let”, “upon”,
and “earth”. The second path is between the words “god”, “saw”, and “good”,
representing a reflection of the creation and a confirmation: “God saw that it was
good”. The third path is between the words “morning, “evening”, and “day”,
representing the cycle of time: “And the evening and the morning were the first/
second/third … day”.

2.4 Identify the central words

Calculating and visualizing the centrality of the words within the network not only
helps to identify the important words in the discourse, but also to highlight the
importance and uniqueness of each cluster. Words with a high degree of centrality
are those that are frequently mentioned with many other words. They help to
identify the main topics around which themes evolve. Words with high
 How to conduct semantic network analysis 23

betweenness are positioned in between other words. Similar to degree centrality,

words with higher betweenness are often also connected to many other words
within the cluster, and therefore share the meaning with other words within it.
They can be helpful to summarize the main idea, frame, theme, and context of
each cluster. But betweenness is measured in relation to all the words in the entire
network. Words with higher betweenness are also located between clusters. As will
be demonstrated throughout the book, these words are therefore useful to extract
the overall narrative of the text. Another advantage of betweenness centrality is
that it is much less equally distributed than degree centrality, and therefore it helps
to visually emphasize key words in the text. The meaning of different centrality
measurements of words in semantic networks is still not entirely clear enough, and
there is room for systematic investigation in this direction. The next section of this
chapter and the following chapters will demonstrate the importance of the cen-
trality measurements, and particularly the importance of betweenness centrality
when conducting semantic network analysis.

2.5 Interpret the network

Perhaps the most crucial and creative stage is the interpretation of the network.
In order to achieve a meaningful analysis, it is often necessary to repeat earlier
stages, merge similar words, remove irrelevant words, less prevalent links, and
even too central words that appear with almost all other words, and thus blur the
network structure. A good method would be to locate about two to seven dis-
tinctive clusters of words that convey different themes and topics, providing a
descriptive map of the content. Interpreting the meaning of central words, pre-
vailing paths or links between words, and network clusters always requires going
back to the text itself, and qualitatively inspecting the context in which they
appear.
It is worth noting that the strength of the semantic network analysis lies in its
ability to extract prevailing themes from large texts (of at least a few thousand
words), and when the chosen words in the network have clear theoretical rele-
vance and meaning. It is less useful to identify esoteric or hidden issues, particu-
larly if the chosen words for network analysis are only based on their frequency.
In the concluding chapter you can find some more detailed guidelines for the
interpretation of semantic networks with examples from the different studies in
the book.
Table 1.1 summarizes the various stages and offers some free available web-based
and software tools as well as R and Python packages that can be used to perform
each stage in the semantic network analysis.
The next sections offer two studies to demonstrate the stages of semantic net-
work analysis and its power to identify the themes and biases of a text. These stu-
dies were chosen as they both follow all the stages mentioned above, but at the
same time differ in terms of the type of source used for the analysis, and the focus
of analysis. While the first example uses news data from Nexis Uni and focuses on
24 E. Segev

a specific issue (the discourse around “fake news”), the second example uses social
media data from Twitter and focuses on the coverage of a specific country (the
discourse on China). In the first example, the analysis unveils the main political
actors and the conflicts of power between politicians, the media, and social media
users. In the second example, the analysis provides a useful starting point to
understand the American-centric views raised by English speaking social media
users, and the potential of Chinese public diplomacy practitioners to alter or at least
provide alternatives to the mainstream discourse.

Example 1: The discourse of elite American newspapers on “fake

news”
The concept of “fake news” has recently become central in the media discourse. It
reflects the era of information overflow and the increasing misinformation and
disinformation, which often spread quickly in social media for various political and
economic reasons. Although deliberate misinformation and propaganda have been
used throughout history, the term “fake news” has become part of the media dis-
course following the 2016 US election, associated with Donald Trump’s strategy to
undermine the news media outlets that criticized him.
The first example of semantic network analysis is based on the study of Segev
and Boudana (2019), and reveals how two elite American newspapers, The New
York Times and The Washington Post, define and frame the emerging concept of
“fake news”. As both sources are widely regarded as credible, the notion of fake
news challenges their core identity and ideology. Professional journalists, who
strive to expose the truth, are threatened by the idea that we live in a post-truth
era, where information is merely about the representation of opinions rather than
the presentation of facts.
In their study Segev and Boudana (2019) collected data from Nexis Uni con-
taining the expression “fake news” in articles that appeared in The New York Times
and The Washington Post between 2016 and 2018 and analyzed about 200 news
items from each outlet. Following the semantic network analysis procedure
described above, in the first stage they identified the most frequent words in the
text and cleaned the list of desirable words (Stages 1.2–1.3). Table 1.2 exhibits this
process, which includes the removal of stop-words and obtaining a list of relevant
frequent words appearing in the context of “fake news”.
The next stage was to obtain a link list of frequent word pairs that appeared
together in the corpus. If the analysis had focused on certain actors, such as country
names or politicians, then the window of their reoccurrence might be the entire
news item. When it comes to frequent words, however, it is recommended to look
at relatively short segments of text, and thus the window in this case would be
their appearance together in the same sentence. When analyzing social media data,
such as tweets or posts, one might prefer to look at co-occurrences of frequent
words within each tweet/post and therefore technically the separation between
windows would be by rows or lines. The link list of all possible pairs of frequent
 How to conduct semantic network analysis 25

TABLE 1.2 Identifying the most frequent and relevant words associated with “fake news”

Word Fre- % Word Frequency % Word Fre- %

quency quency
The 21,910 5.53% Trump 2343 0.59 News 4288 1.08
Of 10,247 2.59% Said 2309 0.58 Fake 2765 0.7
To 10,047 2.54% By 2137 0.54 Trump 2343 0.59
A 9709 2.45% His 1867 0.47 Facebook 1388 0.35
And 8523 2.15% Has 1812 0.46 Media 1334 0.34
In 6685 1.69% Not 1713 0.43 President 1126 0.28
That 6149 1.55% From 1667 0.42 People 1077 0.27
news 4288 1.08% At 1666 0.42 Company 735 0.19
On 3758 0.95% Have 1663 0.42 Election 687 0.17
It 3588 0.91% But 1646 0.42 Social 599 0.15
Is 3515 0.89% They 1596 0.40 Stories 548 0.14
For 3208 0.81% An 1596 0.40 Political 540 0.14
Fake 2765 0.70% Are 1532 0.39 False 526 0.13
He 2783 0.70% About 1506 0.38 Twitter 510 0.13
Was 2458 0.62% Mr 1499 0.38 Online 449 0.11
As 2452 0.62% Be 1491 0.38 Cam- 445 0.11
paign
with 2382 0.60% This 1455 0.37 Fact 428 0.11

words can be then imported to any available network analysis software to obtain
the network graph. In this example, the network analysis and visualization software
Visone (Brandes & Wagner, 2004) was used to construct the semantic network for
further analyses.
Following the procedures discussed above, the next stage would be to sparsify the
network to make it more readable and clearer for interpretations (Stages 2.1–2.2).
Figure 1.4 portrays the process of network sparsification. At first (point “a”), the net-
work is very dense and hard to interpret as it is based on all possible links between the
chosen words, even if they appear together only once. When removing all word pairs
that appear less than 50 times (about 41% of the links), the network becomes much
sparser (point “b”). Still, there are two central actors (the words “fake” and “news”
in black) that appear more frequently with most other words and thus potentially
overshadow the network structure. Since these two words were also used as the
search query to obtain the news corpus, it is reasonable to expect that they would
co-occur with most other words. If the aim of the study is to reveal the themes
related to the fake news discourse, removing them would allow to focus on other
words apart from “fake” and “news” and obtain a cleaner and more structured
network (point “c”).
26 E. Segev

FIGURE 1.4 The process of network sparsification.

Following Stage 2.3, using Louvain modularity (Blondel et al., 2008), which is an
available feature of grouping nodes in many network analysis tools, it is possible to iden-
tify several clusters of words that tend to appear together in the same sentence. These
words often construct common meanings, and thus the cluster analysis stage enables us to
reveal the various themes and frames in the text. The network in Figure 1.5 displays four
distinctive clusters as obtained from this analysis. Cluster “a” at the top left is the largest

FIGURE 1.5 Cluster analysis of the “fake news” discourse in elite newspapers.
 How to conduct semantic network analysis 27

cluster and contains words such as: “media”, “press”, “reports”, “journalists”, “free”,
“speech”, “public”, and “trust”. It clearly represents the media context of the fake news
story, and particularly the important role of journalists. Cluster “b” at the top-center
contains words such as: “people”, “spread”, “believe”, “false”, “stories”, “real”, and
“fact”. This cluster reflects the context of the people or the users and their tendency to
believe in false stories. Cluster “c” at the top right contains the words: “site”, “website”,
“article”, “advertising”, “ads”, and “share”. This cluster deals with the economic and
technological perspective of the story. Finally, the fourth cluster (“d”) at the bottom
includes the words: “election”, “campaign”, “political”, “presidential”, “influence”,
“vote”, “online”, and “information”. This cluster clearly and predominantly deals with
the political aspect of the fake news story.
Figure 1.5 also represents the size of the words based on their betweenness
centrality (indicated in Stage 2.4). As explained above, there are several possible
ways to measure the centrality of words. In this network the rankings of words,
based on their degree and betweenness centrality, strongly correlate (Spearman r =
0.88, p < 0.001). There is therefore no reason to visually distinguish between the
two measurements in the current analysis. In fact, in most cases betweenness cen-
trality is the most useful measurement to highlight the differences between words.
The distribution of betweenness centrality is often much more unequal than that of
degree centrality. When visualizing the size of the words based on the betweenness
centrality, usually only a few words appear to be much more central than others.
Since betweenness measures to what extent each word is positioned in between
two clusters, in most cases each cluster gets one or two central words that often
help us to interpret and contextualize its general meaning (see also the concluding
chapter). Thus, in the following example, the word “media” appears as the most
central in cluster “a”, representing the role of traditional media in fighting fake
news, the word “people” in cluster “b” represents the naivety of the people who
become victims for false information, the word “site” in cluster “c” represents the
platforms and the economic interests, and the word “election” in cluster “d”
represents the political motives behind the fake news phenomenon.
To summarize, based on the network of word co-occurrences in the study of
Segev and Boudana (2019) the fake news discourse as presented in The New York
Times and The Washington Post evolves around the political and economic conflict
between four core players: the media, people, politicians, and commercial compa-
nies. It is described as the attempt of politicians and companies to spread false
information through social networks. While people tend to believe this informa-
tion, the role of traditional media and its journalists is to fight back, present the real
facts, and strive for the truth. It is important to bear in mind that this narrative is
reported by elite American newspapers that attempt to find and sometimes re-
invent their role and identity in this dynamically changing informational world.
Their underlying assumption is that there is a universal concept of truth, and that
they must struggle to unveil and guard it. A whole tradition of journalism and an
emerging business model of premium and quality content depends on their ability
to convince others in this narrative.
28 E. Segev

Example 2: Twitter discourse around “#China”

Social media provide a rich source for content analysis, and Twitter is particularly
accessible for research through its API. Digital diplomacy has embraced Twitter as
one of its prime channels in the last decade, with politicians, state officials, and
journalists sharing their goals and agendas, but also collecting information from
others (Manor et al., 2015). When collecting massive data from Twitter it is diffi-
cult to know exactly who the users are. There are also more and more bots that
produce spam to promote economic and political goals on Twitter and other social
media channels. Still, Twitter is often used by many real people and is considered
as one of the more accessible and useful sources for content analysis.
The study of Chariatte, Segev, and Ingenhoff (2019) demonstrates how semantic
network analysis can be used to learn about the Twitter discourse on China in
English. Given that Twitter is blocked in China, the assumption is that English
tweets would represent Western and predominantly American views, as most
English tweets on China were generated in the US. In their study they analyzed
more than 50,000 tweets for “#China” appearing in January 2019. They gathered
tweets using TAGS v6.1, a freely available extension for Google Spreadsheet that
connects to the Twitter API, and automatically collects streaming data from
Twitter (Hawksey, 2019).
First, following Stage 1.1, Chariatte et al. (2019) removed non-English tweets
and duplicate items to obtain a clean sample of about 15,000 unique tweets.
Although a sample of this size can easily be used for the semantic network analysis,
they carried out their analysis based on a smaller subset of 2500 tweets (about 15%
of the original sample). When comparing the results obtained from several random
subsamples with those of the larger sample, they realized that smaller subsets show
identical patterns. The advantage of using a smaller subset was that it increased the
speed and efficiency of the analysis. This means that despite the availability, Big
Data studies are not always the best solution as they require much more compu-
tational resources and, depending on the research question, they often lead to
similar results. When the purpose of the study is to find the main themes and
frames, a random sample of a few thousand units (be it sentences or posts) would
often be sufficient.
In accordance with Stages 1.2–1.3, Chariatte et al. (2019) identified the most
frequent words, and removed irrelevant words, including the expression “#China”,
which was used as the search query and therefore appeared in all tweets. Figure 1.6
portrays the network of co-occurring words, the main clusters obtained, and the
most central words based on their betweenness centrality (Stages 2.1–2.4).
As can be seen from the cluster analysis using Louvain modularity, the Twitter
discourse on #China revolves around four main themes. Cluster “a” at the top left
includes the central word “travel” with other words such as “warning”, “detained”,
“arrest”, “Huawei”, and “Canada”. It deals with the growing tension over travel
between the US and China. Two international incidents were widely debated by
Twitter users. One is the US travel warning to China, and the other is the arrest of
 How to conduct semantic network analysis 29

FIGURE 1.6 Cluster analysis of the #China discourse in Twitter.

Note. The size of a node indicates its betweenness centrality. The width of the ties
indicates the number of sentences in the corpus that mentioned each pair of words.
Clusters of countries are based on Louvain modularity. Network visualization and cal-
culations were performed with Visone (Brandes & Wagner, 2004).

Huawei’s CFO by Canadian Authorities at the request of the US for alleged vio-
lations of the sanctions on Iran.
Cluster “b” at the top-middle includes many hashtag countries. Looking back at
the content of tweets, these hashtags appear in various contexts and topics ranging
from cryptocurrency to conservative discourse on US international expansion and
control. Cluster “c” at the top right includes words such as “news”, “moon”,
“spacecraft”, and “rover”. This cluster refers to the news on China’s achievement
to land a rover on the far side of the moon. Finally, the largest cluster “d” at the
bottom includes words such as “#apple”, “#economy”, “#trade”, “#trump”, and
“#tradewar”. In this cluster the hashtag #apple gets the highest betweenness cen-
trality, reflecting the company’s efforts to remain dominant in China by skillfully
complying with Trumps’ trade war policy and the Chinese economic ambitions.
In short, the study of Chariatte et al. (2019) showed that Twitter discourse on
China in English was largely news-related. It focused on the political, technologic,
and economic developments of China, and represented almost exclusively the
American point of view. In particular, the content of tweets was primarily driven by
US concerns over China’s economic and technologic expansion, as well as the
political and economic tensions between the two countries. One of the interesting
findings of this study is that there is no counter-narrative to represent the Chinese
point of view. A possible application of this study is in the field of public diplomacy.
In order to improve its global image, China could employ popular social media
channels such as Twitter and produce English content to present alternative views.
30 E. Segev

Conclusion
Texts are the prime source of analysis in many political, social, and communication
informed studies. This chapter introduced a new emerging approach, semantic network
analysis, to reveal the main themes and biases in any given text. The advantages of this
approach are numerous, particularly today when information is easily available, and data
science is in demand. Still, there are not many studies to date that employed semantic
network analysis in the social sciences and particularly in the field of communication
and politics. Moreover, the few studies that focus on the process of semantic network
analysis tend to be technical and require some previous knowledge of programming and
network analysis principles. The purpose of this chapter was to offer some basic guide-
lines and steps to conduct semantic network analysis, to offer some free available tools,
and to demonstrate how it can be applied without previous knowledge and experience.
The main stages of semantic network analysis include choosing and preparing the
sample for analysis, selecting and refining the list of desired words to be included in
the analysis, converting the list of words into a link list format and importing it to
network software, sparsifying and cleaning the network for analysis, and finally,
identifying clusters and central words. Once the network is readable, the linear text
turns into a two-dimensional descriptive map, which allows to identify the main
themes and even the biases and underlying narratives. Although this process can take
some time, it is still much faster than traditional qualitative and quantitative methods
of content analysis. Moreover, it is fit for small and large texts and therefore ideal for
mapping the content in Big Data studies. The applications of semantic network
analysis as discussed in the introduction to this book are diverse and can support all
stages of the research process, starting from the mapping of literature, developing
theories from the bottom-up, to testing existing theories and hypotheses.
Finally, the potential of semantic network analysis was demonstrated in two previous
studies. The first was a study of Segev and Boudana (2019), who mapped the discourse
around “fake news” in The New York Times and The Washington Post. After identifying
the most relevant frequent words and constructing networks of word co-occurrences,
they found four main clusters, reflecting the main narratives of elite American news-
papers around the fake news discourse. The second example was based on the study of
Chariatte et al. (2019). In their study they employed Twitter API to learn about the
social media discourse around China in English. They showed that news reports about
China tremendously influence the discourse on Twitter. They also found that Amer-
ican views dominate this discourse, reflecting US concern over China’s economic
expansion and the political and economic tensions between the two countries.
To conclude, semantic network analysis is a powerful and promising tool for
researchers to identify patterns in any text, be it books, academic papers, social media,
songs, films, or even interview transcripts. One way to look at it is as an x-ray of text,
as it assists scientists to unveil the structure of content, the relationships between words,
and their overall position. By offering available online tools, simple procedures, and
concrete case studies, this book hopes to encourage more researchers to engage in
semantic network analysis and make it an essential part of the students’ curriculum.
 How to conduct semantic network analysis 31

References
Bail, C. A. (2016). Combining natural language processing and network analysis to examine
how advocacy organizations stimulate conversation on social media. Proceedings of the
National Academy of Sciences, 113(42), 11823–11828.
Benoit, K., Watanabe, K., Wang, H., Nulty, P., Obeng, A., Müller, S., & Matsuo A.
(2018). quanteda: An R package for the quantitative analysis of textual data. Journal of
Open Source Software, 3(30), 774.
Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding
of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment,
10, 10008.
Brandes, U., &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In
M. Jünger & P. Mutzel (Eds.), Graph Drawing Software (pp. 321–340). New York:
Springer-Verlag.
Chariatte, J., Segev, E. & Ingenhoff, D. (2019). Tweeting #Country: A comparative analysis of
country images as a new instrument for developing international communication strategy.Presented at
The Annual Conference of the International Association for Media and Communication
Research, IAMCR, 7–11 July 2019, Madrid, Spain.
Danowski, J. A. (1993). Network analysis of message content. In W. D. Richards & G. A.
Barnett (eds.), Progress in Communication Sciences, Vol. 12 (pp. 198–221). Norwood, NJ:
Ablex Publishing Corp.
Drieger, P. (2013). Semantic network analysis as a method for visual text analytics. Procedia—
Social and Behavioral Sciences, 79, 4–17.
Hawksey, M. (2019). TAGS. Twitter archiving through Google Sheets. https://tags.ha
wksey.info.
Higuchi, K. (2001). KH Coder: Free software for quantitative content analysis or text mining. https://
khcoder.net.
Lindner, G., Staudt, C. L., Hamann, M., Meyerhenke, H., & Wagner, D. (2015, August).
Structure-preserving sparsification of social networks. 2015 IEEE/ACM International Conference
on Advances in Social Networks Analysis and Mining (ASONAM) (pp. 448–454). New
York, IEEE.
Manor, I., Segev, E., & Kampf, R. (2015). Digital diplomacy 2.0? A cross-national com-
parison of public engagement in Facebook and Twitter. The Hague Journal of Diplomacy,
10(4), 331–362.
Paranyushkin, D. (2011). Identifying the pathways for meaning circulation using text network
analysis. Nodus Labs, 26. https://noduslabs.com/publications/Pathways-Meaning-Text-Net
work-Analysis.pdf.
Rule, A., Cointet, J. P., & Bearman, P. S. (2015). Lexical shifts, substantive changes, and
continuity in State of the Union discourse, 1790–2014. Proceedings of the National Academy
of Sciences, 112(35), 10837–10844.
Segev, E. (2020). Text and network analysis tools. http://www.eladsegev.com/tools.
Segev, E. & Boudana, S. (2019). When fake news makes the news: Definitions of fake news in american
newspapers.Presented at the International Conference of the European Forum on Fake News
and Disinformation. May 15, 2019, The Hebrew University of Jerusalem, Israel.
Thelwall, M. (2018). Social Web Text Analytics with Mozdeh. Wolverhampton: University of
Wolverhampton.
Van Ham, F., Wattenberg, M., & Viégas, F. B. (2009). Mapping text with phrase nets. IEEE
Transactions on Visualization and Computer Graphics, 15(6), 1169–1176.
2
THE NEWS COVERAGE OF THREATS
Iranian nuclear programs in Israeli press

Noa A. Hatzir, Elad Segev, Kohei Watanabe and Atsushi

Tago

Introduction
International news often deals with conflicts, and global nuclear threats have been
at the top of the international agenda. The news coverage of Iran is particularly
prominent and can roughly be divided into before and after the Islamic revolution
of 1979. In general, Western views on Iran before the revolution were rather
positive due to its pro-Western and secular monarchic system. Iran’s nuclear pro-
gram began in the 1950s with the help of the US and other Western countries as
part of the Atom for Peace program. In 1970 Iran ratified the treaty on the Non-
Proliferation of Nuclear Weapons (NPT). After the revolution and during the war
with Iraq, the Iranian economy struggled, and nuclear development was reduced
until the 1990s. Then, with support from Russia, the nuclear program was taken
up again, and the Bushehr nuclear plant was finished.
In the last two decades there has been an ongoing tension between Iran and the
West as Iran has repeatedly violated its commitments to the NPT. The IAEA (Inter-
national Atomic Energy Agency) revealed that Iran’s nuclear program has rapidly
increased, with two sites under construction that were not declared. As a result of
international pressure, the “Teheran Declaration” was introduced in 2003, demanding
Iran suspend its uranium enrichment activities. In return, EU3 (UK, France, and
Germany) agreed to recognize the rights of Iran to peaceful use of nuclear energy by
following the NPT guidelines. The negotiations lasted until 2005 but failed, mainly
due to the American administration’s insistence on “zero enrichment” (Parsi, 2013).
This contributed to the victory of Mahmoud Ahmadinejad, who criticized “weak”
Iranian foreign policy (Kazemzadeh, 2017).
After Ahmadinejad took office, Iran resumed its enrichment program at a few
sites, rejecting IAEA resolutions and postponing international negotiations, arguing
that foreign sanctions do not affect Iran (Mousavian & Mousavian, 2018). This

DOI: 10.4324/9781003120100-3
 The news coverage of threats 33

policy lasted until 2013, when Hassan Rouhani, who was a chief negotiator in the
“Teheran Declaration”, was elected as president. Rouhani resumed the negotia-
tions with the West, wishing to continue nuclear development for energy purposes
and preserve Iran’s diplomatic relationships (Mousavian & Mousavian, 2018). This
change in office raised hopes among the Iranians and the outside world to again
achieve a mutual trust (Yazdani, 2019).
In 2015, Iran signed the Joint Comprehensive Plan of Action (JCPOA) (Alarqan,
2020; Mousavian & Mahmoudieh, 2017; Pieper, 2019; Yazdani, 2019) with the P5+1
(consisting of the UK, France, Germany, China, Russia, and the US). This agreement,
however, did not last long. In May 2018, the US withdrew from JPCOA upon pres-
sure by Israel and Saudi Arabia, and sanctions were imposed on Iran. Rouhani decided
that Iran would stick to the terms of the agreement with the other signatories, the
European countries, China, and Russia (Alarqan, 2020; Pieper, 2019). Yazdani (2019)
argues that despite the successful conclusion of the nuclear deal, Iran’s political and
economic ties with the world did not recover after the 1979 Islamic revolution.

Israel’s position on the Iranian nuclear development

Before the Islamic revolution, the economic and diplomatic relationship between Iran
and Israel flourished. Part of it was because they shared a common threat, being sur-
rounded by Arab and Sunni countries. Since the Islamic revolution, and as the Cold
War resumed, this special relationship dramatically changed (Maher, 2020). Nowa-
days, the Israeli government perceives Iran as “committed ideologically, officially, and
practically to the destruction of Israel” (Danieli, 2019, p. 129), mainly because Iran
does not recognize Israel as a state and develops nuclear weapons.
Following this view the Israeli political and military elite emphasize that the
Iranian threat to Israel is existential (Merom, 2017). It is reflected by the media and
several politicians, such as the Israeli Prime Minister, Benjamin Netanyahu, linking
the Iranian nuclear threat with the Jewish Holocaust. It is materialized today
through the economic and military support of Iran to Hamas and Hezbollah,
organizations that continuously fight against Israel.
Hence, politicians and military officials often express skeptical views on any inter-
national attempt to negotiate with Iran, stressing the need to prevent Iran from
developing a nuclear program. Their assumption is that the Iranian regime is zealous
and irrational (Merom, 2017), and Israeli national intelligence continuously reveals its
violations of international commitments (Danieli, 2019; Jones, 2018). Israel’s policy is
therefore aligned with that of the US under Trump’s administration and Saudi Arabia
(Maher, 2020). The harsh opposition of Israel and the US regarding the Iranian
nuclear program raises criticism among some observers. The question is whether a
country is legitimized to request another country to stop the development of nuclear
power, while disposing of nuclear facilities and not being part of the NPT (Danieli,
2019; Pedraza, 2017; Seybert, 2019). Israel has been ambiguous about its own nuclear
capacities (Pedraza, 2017; Seybert, 2019), but at the same time emphasizes the need to
secure its existence and the security of its citizens (Pedraza, 2017; Seybert, 2019).
34 Hatzir, Segev, Watanabe, and Tago

The coverage of international conflicts in the news

When it comes to the coverage of international conflicts it is well known that the
media often rely on official government sources and are aligned with national agen-
das (Bennett, 1990; Livio & Cohen-Yechezkely, 2019). One reason for this could be
the lack of knowledge about most foreign countries and about international relations.
When covering security issues, journalists often depend on government sources for
gathering security-related stories (Farnsworth et al., 2010), having limited, if any,
access to the covered zone (Livio & Cohen-Yechezkely, 2019). The lack of alter-
native sources on security issues allows governments to control the flow of infor-
mation and overplay the level of security threat in order to gain public support for
their foreign policies. Hence, national outlets generally offer similar interpretations
and frames (Bennett, 1990), which support the views and actions of the official
authorities with hardly any criticism (Feinstein, 2016, 2018; Karniel et al., 2017;
Yarchi et al., 2015; Randahl, 2018). In particular, supportive news coverage occurs
in times of national threat (Lavie-Dinur et al., 2018; Randahl, 2018; Yarchi et al.,
2015) or during a war (Livio & Cohen-Yechezkely, 2019; Feinstein, 2018), when
journalists are much more committed to the state’s ideology (Liebes, 1997). In those
moments, media frames emphasize much more the distinction between “us”, the
good, and “them”, the bad (Carter et al., 2011).
Patriotic media frames were found in news coverage in Israel (Liebes & Kampf,
2009; Karniel et al., 2017). Different media outlets displayed similar national nar-
ratives when covering wars, conflicts (Livio & Cohen-Yechezkely, 2019), and
terror attacks (Lavie-Dinur et al., 2018). Still, several differences were found even
when reporting on terror attacks, reflecting the newspaper’s political inclination.
Thus, for example, Haaretz, an elite leftist newspaper presented a complex story
and attempted to blur the dichotomy between “us” and “them”. Ynet and Walla,
popular outlets with a center-right inclination, focused mainly on the terror attacks
and displayed a much simpler story with a clear distinction between the good and
the bad (Lavie-Dinur et al., 2018).
At the same time, studies show that journalists do not completely adjust to the
governmental narrative and some attempt to maintain their own voices. This could
happen, for example, when journalists present the other side of the story, the counter
perspectives (Althaus, 2003; Entman, 2003; Hayes & Guardino, 2010; Kampf &
Liebes, 2013; Neiger et al., 2010). Critical voices can also be found in the stance or
tone of reporting (Neiger et al., 2010) and cynical expressions (Cappella et al., 1997).
In times of conflict, self-reflection is especially important, but also challenging, as it
exposes the weaknesses of the nation and encourages reactions of the enemies.

The news coverage of nuclear threats

The international debate on nuclear weapons emerged after the Second World
War with the creation of UN forums and conventions on this topic (MacDonald &
Hunter, 2019). In 1957 the International Atomic Energy Agency (IAEA) was
 The news coverage of threats 35

established to balance, supervise, and control nuclear technology, making sure it is

used as a practical and useful tool rather than as a weapon (IAEA, 2020). The first
convention, the Non-Proliferation Treaty, was signed by 195 countries in 1968 as
part of a global effort to prevent the development of mass destruction weapons and
promote peaceful uses of nuclear energy (UN, 2020).
Together with diplomatic efforts, the nuclear program in Iran has attracted media
attention around the world (MacDonald & Hunter, 2019; Segev, 2016). The framing
of nuclear power, however, has changed over the years, with nuclear energy being a
symbol of progress (Gamson & Modigliani, 1989), but at the same time dangerous
(such as in the case of the Chernobyl or Fukushima nuclear accidents). Nowadays, the
development of nuclear weapons is mainly discussed in Western outlets in the context
of Iran and North Korea (MacDonald et al., 2015; MacDonald & Hunter, 2019).
Most of the news discourse is simplistic, portraying Iran (Atai & Mozaheb, 2013;
Hearns-Branaman, 2017; KhosraviNik, 2015a, 2015b), North Korea (Lim, 2014), and
their respective supporters, Russia and China, as the provocateurs (Boudana & Segev,
2017) that threaten the peace and security of the “good” Western countries.
Izadi and Saghaye-Biria (2007) further reveal that American newspapers describe
Iran as a fundamentalist Muslim regime, and a significant threat to the US and the
world. The Iranian government is characterized as irresponsible and not trustable,
especially when dealing with nuclear technology. It is also found that American
newspapers rely on official US and Israeli sources and get to the conclusion that
Iran develops a clandestine nuclear weapons program that is not under interna-
tional inspection (Izadi & Saghaye-Biria, 2007). British (KhosraviNik, 2015b),
American (Amin, 2019; Izadi & Saghaye-Biria, 2007; Kadkhodaee & Ghasemi
Tari, 2019), and Israeli (Klein, 2009) newspapers covering the Iranian nuclear
program adopt the rhetoric of “them”, the Iranian, and “us” the Western coun-
tries, with Israel and Arab states being the victims.
In the Iranian media a parallel rhetoric was found. The Western countries are
represented as evil (KhosraviNik, 2015b) and Israel is represented as an opponent
(Klein, 2009) and as a terrorist threat (Jaspal, 2014). In fact, unlike most Arab media,
the Iranian media keep referring to Israel as the “Zionist regime” to underline that
the Iranian government does not recognize it. In contrast to Western media, the
Iranian newspapers display the nuclear program as peaceful (Ghane & Mahdavirad,
2016; Klein, 2009).
Framing the “other” as a threat is a common mechanism to increase national
identity and legitimation of one-self (Amin, 2019; Jaspal, 2015; KhosraviNik, 2015b;
Klein, 2009; Nossek, 2004), alongside with developing perspectives of power and
control (Amin, 2019). Moreover, it raises emotions of danger, enmity, fear, and
anger. The notion of threat, particularly in the US, is strongly associated with “ter-
rorism”, “radical Islam”, and “nuclear technology” (Woods, 2011). It can be
explained as the psychological responses to danger: dread, risk, judgment, and worry
(Slovic, 2004). Von Sikorski et al. (2017) warn that the association between terrorism
and radical Islam results in “negative perceptions of Muslims in general and may
further enhance intergroup conflicts between Muslims and non-Muslims”.
36 Hatzir, Segev, Watanabe, and Tago

Klein (2009) compared the coverage of the Iranian nuclear program in Israel and
Iran. He believes that they both attempt to “bridge” their interpretations of events
to their allies. The Israeli newspaper Ynet presented the worst-case scenario for
Israel: the accomplishment of Iranian atomic weaponry. This scenario was pre-
sented as a threat not only to Israel but also to the entire world, thus bridging the
threat to its Western allies. On the other hand, the Iranian newspaper IRNA
underscored the peaceful nuclear activities by encouraging its allies to continue and
support its nuclear program.
In short, nuclear development in Iran has been observed with great suspicion
among Western countries. It is framed as a world threat that can potentially escalate
into a war, blaming the fundamentalist and irresponsible Muslim regime in Iran.
Although the news coverage of this crisis is critical to the stability of the region and
the entire world, there are very few studies to date that looked at the coverage in
Israeli media. Most studies on Israeli media focused on wars and terror attacks
within or along its borders.
It is important to understand the frames offered by the Israeli media for several
reasons. First, unlike terror attacks and other regional conflicts that appear in waves,
the Iranian nuclear threat is an ongoing conflict that influences people’s perceptions
in the long term. Second, even though the threat is not geographically close, it
affects the daily lives of Israeli citizens and is part of the daily discourse. Third, a
war against Iran will have consequences for the entire world. Studying the way one
country perceives and frames the other is often a starting point for dialogue and
peaceful diplomacy.

Methods

Sample
The corpus of our study included all the print news articles in the two leading
Israeli newspapers, Yedioth Ahronoth and Haaretz, that were published between
2008 and 2018 and mentioned the word “Iran” in Hebrew (“‫)”איראן‬. In total, we
retrieved N = 8779 items from the archives of Yedioth Ahronoth and N = 15,930
items from Haaretz.

Choosing relevant words

First, we generated a list of word frequencies, removing stop-words and irrelevant
words (Segev, 2020). To determine the relevance of the words we manually
inspected the context in which the words appeared in the text. Following groun-
ded content analysis procedures, we manually classified words that provided clear
information about the Iranian–Israeli conflict. This classification partly corre-
sponded to Entman’s (2007) conflict frames (problem, cause, judgment, solution),
and partly derived from the grounded content procedure. This process yielded
seven general categories of words based on their functions in the conflict process:
 The news coverage of threats 37

actors (e.g., country names, organizations, and politicians), assessment of the situation
(e.g., threat, crisis, or war vs. security or peace), stand or position (e.g., extreme or
hostile vs. moderate or friendly), intentions or interests, actions (e.g., attack or
bombing vs. negotiation or agreement), resources (e.g., weapon vs. energy), and
finally, outcomes or results (e.g., failure vs. success).
Obviously, the context of the coverage of Iran in Israel is predominantly negative.
Still, a careful inspection of the text also helped us to classify whether some of these
words are related to the threat dimension and the escalation of the conflict (e.g.,
“danger”, “war”, “attack”, “missiles”) or rather to the opportunities for de-escalation
of the conflict and the steps required for regional stability (e.g., “security”, “peace”,
“negotiation”, “diplomacy”). In the final analysis we only included words with a
frequency of 0.01% of the entire corpus and above, resulting in a list of the top 268
frequent words for Haaretz and 224 words for Yedioth Ahronoth. We did not use the
same word list for the comparison as each outlet may have different emphasis or even
different frames. Still, 194 words (72% of the words in the list of Haaretz and 87% of
the words in the list of Yedioth Ahronoth) were identical in both newspapers.

Construction and analysis of semantic networks

Based on the chosen frequent words we constructed a network of word pairs appear-
ing in the same sentence (269,324 sentences for Yedioth Ahronoth and 342,262 sen-
tences for Haaretz). Following the procedure described in earlier chapters, we
sparsified the semantic networks by removing the words “Iran” (the search term itself)
and “Israel”, as both appeared at the absolute center of the network and prevented the
identification of the overall network structure and the presence of word clusters. We
also deleted less frequent links between word pairs (lower than 20 mentions in Haaretz
and 10 mentions in Yedioth Ahronoth). This process yielded much clearer and readable
networks for visualization and interpretation purposes. Finally, we calculated some
important centrality measurements (including degree centrality and betweenness) and
employed Louvain modularity (Blondel et al., 2008). As was explained earlier, this
method is particularly useful to identify clusters of words that tend to appear together
in the same sentence. The result section below outlines the networks of actors
involved in the coverage of Iran (politicians, organizations, and countries) as well as the
networks of topics and issues (actions, opinions, and views).

Results

Actors related to the coverage of Iran in Israeli newspapers

Figure 2.1 portrays the network of the most frequent actors mentioned in the news
about Iran as appeared in Haaretz between 2008 and 2018. The cluster analysis
using Louvain modularity identified six groups, which can be clearly interpreted as
corresponding to six countries or regions that are involved in the development of
nuclear power in Iran. The cluster on the top-left shows the Israeli Prime Minister
38 Hatzir, Segev, Watanabe, and Tago

FIGURE 2.1 Network of co-occurring actors in Haaretz news coverage on Iran.

Note. Words are translated to English. The size of the nodes reflects their relative
betweenness level, and the width of the ties the number of sentences that mentioned
each pair of words together. Word clusters are based on Louvain modularity (Blondel et
al., 2008). The network layout applies stress minimization to a metric MDS. Network
visualization and calculations were performed with Visone (Brandes & Wagner, 2004).

Netanyahu and two heads of two right-wing parties, Naftali Bennett and Avigdor
Lieberman. Further clockwise is the Palestinian Authority and Hamas. The next
cluster on the top-right contains Syria with some links to the Iranian Revolu-
tionary-Guards, Lebanon, and Turkey. On the bottom-right is Egypt, with links to
other Middle Eastern countries, notably Yemen, Jordan, Qatar, and Saudi Arabia.
The largest cluster by far at the bottom-center is that of the US, mentioning the
presidents Barack Obama and Donald Trump. Interestingly, Kim Jong-il, the
North Korean leader, is also mentioned in relation to Trump, marking the con-
nection between the Iranian and the North Korean nuclear threat to the West.
Finally, the cluster on the left shows Russia and its President Putin, with some links
to China, Germany, and France.
What is interesting to observe in Figure 2.1 is the presence of political leaders in
the US, mostly located on the left part of the figure, and, on the contrary, the lack
of personalization in the representation of the Arab states on the right. Another
notable pattern is the strong ties between the leaders of Israel and the US, on the
one hand, and Russia and Syria, on the other. This map clearly presents the
polarity between two global powers with interests in this conflict: the US and
Russia, with Israel, Syria, and Iran being their proxies.
The network of co-occurring words in Haaretz, a left-wing newspaper in
Israel, portrays a clear bias toward the Israeli narrative with some limited
 The news coverage of threats 39

representation of Palestine and the Arab world. The network of words in

Yedioth Ahronoth, a centrist newspaper, reflects this biased narrative even more.
Figure 2.2 outlines the main actors involved in the coverage of Iran in Yedioth
Ahronoth (excluding the words Iran and Israel). Like the previous map, there is
a detailed representation of Israeli parliament members (top-right cluster), but
there is an additional cluster representing officials of the Israeli army (top clus-
ter). In line with Haaretz’s narrative, there is a strong presence of American
actors (bottom cluster) and some presence of Russia, China, and European
countries, representing the international community with their various stakes
and agendas.
The remarkable difference between the two outlets is that the representation
of the Arab worlds in Yedioth Ahronoth is even more limited. The cluster of
Palestine, which is significant in Haaretz, does not exist in Yedioth Ahronoth.
The two clusters of Syria and Middle Eastern countries in Haaretz are merged
into one cluster representing the Arab world. Interestingly, Russian President
Putin figures in this cluster rather than in the Russian one, indicating his strong
ties and influence in the region. Overall, while news reports in Haaretz attempt
to give a stage for some Arab voices, in Yedioth Ahronoth the focus of the story
is mostly on Israel and its American ally with a minimal reference to the Arab
world.

FIGURE 2.2 Network of co-occurring words in Yedioth Ahronoth reports on Iran.

Note. Words are translated to English. The size of the nodes reflects their relative
betweenness level, and the width of the ties the number of sentences mentioned each
pair of words together. Word clusters are based on Louvain modularity (Blondel et al.,
2008). The network layout applies stress minimization to a metric MDS. Network
visualization and calculations were performed with Visone (Brandes & Wagner, 2004).
40 Hatzir, Segev, Watanabe, and Tago

Topics and issues related to the coverage of Iran in Israeli newspapers

Figure 2.3 illustrates the main topics and issues covered by Haaretz in relation to the
coverage of Iran. Triangular nodes represent threat-related words, while square nodes
represent opportunity-related words. At the center of the network is the word “war”,
surrounded by words such as “struggle”, “fight”, “enemy”, and “fear”, which together
represent the main threats and fears posed by Iran. On the right is a cluster of the actual
conflict and the arms involved containing words such as “missiles”, “weapon”,
“rockets”, “terrorism”, and “attack”. This implies that the conflict between Israel and
Iran is indirect, manifested by a more direct conflict with Hezbollah, an organization
funded by Iran that operates in Syria and Lebanon.
On the left side, on the other hand, are two clusters representing the “chances”
and hopes for “peace” as a result of “negotiations”, “talks”, and “agreements”.
These clusters also include words such as “uranium enrichment” and “nuclear”,
suggesting that the results of “negotiations” and “sanctions” may offer a possible
solution as they aim at stopping nuclear development. In a way, the cluster on
right side of Figure 2.3 embodies the emotional or the threat dimension in media
coverage, while the left clusters represent the rational or the opportunities dimen-
sion to resolve the conflict.

FIGURE 2.3 Network of co-occurring non-actor words in Haaretz reports on Iran.

Note. Words are translated to English. The size of the nodes reflects their relative
betweenness level, and the width of the ties the number of sentences that mentioned
each pair of words together. Triangular nodes reflect threat context, square nodes
opportunity context, and circle nodes neutral words. Clusters are based on Louvain
modularity (Blondel et al., 2008). The network layout applies stress minimization to a
metric MDS. Network visualization and calculations were performed with Visone
(Brandes & Wagner, 2004).
 The news coverage of threats 41

Finally, the two clusters on the top and bottom highlight the political and economic
aspects of the story. The top cluster focuses on “election” and “demonstrations”,
which mainly appear in the context of the Iranian public and their disagreement with
their hard-line leaders. The cluster at the bottom focuses on the economic aspect with
the word “oil”, surrounded by words such as “stop”, “prevent”, “sanctions”, and
“succeed”. This cluster provides further support to the notion that the Iranian nuclear
threat is also about economic resources and oil, and therefore sanctions and pressure
from the west may bring about a possible solution.
Overall, the main narrative told by Yedioth Ahronoth seems to be similar. Figure 2.4
presents the network of co-occurring words related to Iran in Yedioth Ahronoth after
excluding actor-related words such as politicians, country names, and organizations.
In line with Haaretz, the threat of war is at the center of the story in Yedioth Ahro-
noth. But unlike Haaretz, in which the word “war” is surrounded by international
relation concepts such as “strategic”, “policy”, “relations”, “economy”, or “alliance”,
in Yedioth Ahronoth the word “war” is surrounded by more words referring to actions
such as “prevent”, “succeed”, “damage”, “harm”. The problem as framed by both
newspapers is therefore similar—the threat of war between Israel and Iran, yet the
solution or outcomes are framed differently in each newspaper.

FIGURE 2.4 Network of co-occurring non-actor words in Yedioth Ahronoth reports on Iran.
Note. Words are translated to English. The size of the nodes reflects their relative
betweenness level, and the width of the ties the number of sentences mentioned each
pair of words together. Triangular nodes reflect threat context, square nodes opportu-
nity context, and circle nodes neutral words. Clusters are based on Louvain modularity
(Blondel et al., 2008). The network layout applies stress minimization to a metric MDS.
Network visualization and calculations were performed with Visone (Brandes &
Wagner, 2004).
42 Hatzir, Segev, Watanabe, and Tago

Haaretz emphasizes international relations strategies and diplomacy, while Yedioth

Ahronoth focuses on specific actions such as preventing nuclear development and
protecting Israel’s rights for security. Nuclear development appears in Yedioth
Ahronoth as a separate and distinctive theme (the cluster at the bottom), while in
Haaretz it is strongly related to the agreement talks cluster and thus also for the
opportunity theme. It is worth mentioning the word “Holocaust”, which is con-
nected to the word “speech”. It refers to speeches of the Israeli Prime Minister,
Benjamin Netanyahu, who constantly uses the rhetoric of fear and compares the
Iranian threat to the Jewish Holocaust (see also Chapter 5 for a full analysis of this
phenomenon).
Further, in line with Haaretz, a prevailing theme in Yedioth Ahronoth is the mis-
sile attacks on Israel (bottom-right cluster), which are related to the Iranian-funded
Hezbollah in Syria and Lebanon. Finally, the economic theme (top-right cluster
with words such as “oil” and “suctions”) and the political theme (top-left cluster
with words such as “elections”, “campaign”, and “coalition”) are almost identical
to the themes covered in Haaretz.
In short, it seems that the two outlets generally agree on the main issue. They
both emphasize the narrative of the threat of war between Israel and Iran. Yet the
network of words and the emergence of topical clusters reveal some disagreement
regarding possible solutions.
Another way to present the clusters of words or the common meaning they
share is based on a table. Unlike networks, tables cannot portray the relative loca-
tion of each cluster compared to others, but they can offer a more detailed view of
the composition of words in each cluster, their exact level of network centrality,
and their other properties. As the overall network structure in both outlets was
relatively similar, we present here the analysis based on Haaretz only, which offered
a more detailed coverage and a greater diversity of topics.
Table 2.1 summarizes the list of frequent words in Haaretz divided by their
assigned clusters in the network (based on Louvain modularity), their network
centrality (betweenness and degree in percentages), their role or function (actors,
assessment of the situation, stand or position, intentions or interests, actions,
resources, and outcomes or results), and their tone (threat or opportunity). The last
two categories were assigned after manually inspecting the actual use of each word
in the text. Words with the highest betweenness are marked in bold, reflecting the
main theme in which each cluster of words is mentioned.
Cluster 1 is located at the center as it captures the essence of the Iranian story and
the deepest fear of the Israeli public that the development of nuclear power will
escalate into a war. Together with threat-related words such as “fight” and “struggle”
there are also some opportunity-related words, such as “visit”, “trade”, and “defense”,
which reflect different diplomatic, economic, and military strategies to cope with this
international threat. Cluster 2 focuses on arms and weapons (resources) and therefore
also on actions such as “attack”, “fighting”, and “shooting”, which result in “damage”
and “harm”. Cluster 3 is built around the words “oil” and “money” resources. Action
words such as “prevent”, “stop”, and “sanctions” may lead to the successful
TABLE 2.1 Centrality and clusters of words in Haaretz

Lou- Source between- degree (%) Category Sentiment Louvain Source between- degree (%) Category Sentiment
vain ness (%) cluster ness (%)
cluster
1 War 18.42 7.83 situation Neg 4 Peace 12.91 5.22 situation pos
1 Relations 1.96 3.21 position 4 Solution 1.36 1 Result pos
1 Policy 1.72 3.21 Action 4 Settlements 1.26 0.6 Action neg
1 Struggle 1.26 0.6 Action Neg 4 Good 0.39 1.2 situation pos
1 Economy 1.02 2.61 resources 4 Opportunity 0.01 0.6 situation pos
1 Defence 0.03 0.6 Action Pos 4 Enterprise 0 0.2 Action pos
1 Strategic 0.01 1.2 intentions 4 Building 0 0.2 Action neg
1 Trade 0 0.8 Action Pos 4 Negotiation 0 0.2 Action pos
1 Control 0 0.2 Action 4 Occupation 0 0.4 Action neg
1 Fight 0 0.2 Action Neg 4 Chance 0 0.2 Result pos
1 Visit 0 0.2 Action Pos 4 Danger 0 0.2 situation neg
1 Criticism 0 0.8 Action Neg 4 Problem 0 0.2 situation neg
1 Alliance 0 0.4 position 4 Conflict 0 0.4 situation neg
1 Enemy 0 0.2 position Neg 5 Agreement 15.8 7.03 Action pos
1 Fear 0 0.2 position Neg 5 Talks 4.87 2.41 Action pos
1 Future 0 1 Result 5 Nuclear 3.23 3.41 resources neg
1 Crisis 0 0.4 situation Neg 5 Produce 1.49 1.2 Action neg
2 Missiles 4.83 3.61 Resources Neg 5 Exchange 0.54 1.61 Action pos
2 Weapon 3.06 3.21 Resources Neg 5 Nuc.-prog. 0.52 1.2 Action neg
2 Attack 2.71 2.81 Action Neg 5 Reactor 0.08 0.8 resources neg
The news coverage of threats 43

(Continued)
TABLE 2.1 Cont.

Lou- Source between- degree (%) Category Sentiment Louvain Source between- degree (%) Category Sentiment
vain ness (%) cluster ness (%)
cluster
2 Terrorism 2.64 3.01 Action Neg 5 Transaction 0 0.4 Action pos
2 Security 2.58 4.02 Situation Pos 5 Discussion 0 0.2 Action pos
2 Rockets 1.6 1.41 Resources Neg 5 Proposal 0 0.2 Action pos
2 Threat 0.8 2.61 Situation Neg 5 Progress 0 0.4 Action
2 Harm 0.16 1 Result Neg 5 Contacts 0 0.2 Action pos
2 Information 0.07 0.8 Resources 5 Promised 0 0.2 Action pos
2 Planes 0.04 0.6 Resources 5 Reconcilia- 0 0.2 Action pos
44 Hatzir, Segev, Watanabe, and Tago

tion
2 Targets 0 0.8 Action Neg 5 Uran.-enrich. 0 0.2 Action neg
2 Interests 0 0.8 Intentions 5 Confronta- 0 0.4 Action neg
tion
2 Fighting 0 0.4 Action Neg 5 Resistance 0 0.6 Action neg
2 Warned 0 0.4 Action Neg 5 Assistance 0 0.2 Action pos
2 Shooting 0 0.2 Action Neg 5 Appointment 0 0.4 Action pos
2 Damage 0 0.2 Action Neg 5 Uranium 0 0.6 resources neg
2 Battle 0 0.4 Action Neg 5 Bomb 0 0.2 resources neg
2 Reaction 0 0.4 Action Neg 5 Results 0 0.2 Result
2 Operation 0 0.4 Action Neg 6 Opposition 2.58 1 Position
2 Extreme 0 0.2 Position Neg 6 Elections 2.36 2.41 Action
3 Oil 4.01 2.01 Resources 6 lefts 1.52 0.8 Position Pos
Lou- Source between- degree (%) Category Sentiment Louvain Source between- degree (%) Category Sentiment
vain ness (%) cluster ness (%)
cluster
3 Prevent 1.12 1.41 Action Pos 6 Demonstra- 1.26 0.4 Action Neg
tions
3 Sanctions 1.08 2.41 Action Pos 6 Speech 0.16 0.6 Action
3 Convince 0.3 0.6 Action Pos 6 Support 0.11 0.8 Action
3 Price 0.03 0.6 Result Neg 6 Law 0.06 0.8 Action
3 Succeed 0.01 0.4 Result Pos 6 Division 0.04 0.4 Position Neg
3 Stop 0 0.2 Action Pos 6 Protest 0 0.2 Action Neg
3 Million 0 0.2 Resources 6 Protect 0 0.2 Action Pos
3 Dollar 0 0.2 Resources 6 Democracy 0 0.4 Result Pos
The news coverage of threats 45
46 Hatzir, Segev, Watanabe, and Tago

“reduction” of the oil “prices”. This cluster contains words that are mainly related to
the opportunities for Israel.
The words in Cluster 4 are all about seeking for a peaceful solution. As part of
the opportunity dimension, action-related words such as “negotiation” can poten-
tially increase the chances for peace. At the same time, peace in the Israeli context
is always discussed in relation to Palestine as well. Hence, as can be seen in Cluster
4, threat-related words such as the “danger” of escalation of this “conflict” are
linked with “building” in the “settlements” and “occupation”. This point of view
is specific for Haaretz, a left-wing newspaper, and is completely missing in the
peace-related word cluster of Yedioth Ahronoth.
The words in Cluster 5 further deal with the “progress” of “uranium enrich-
ment” and the development of a “nuclear bomb”. Naturally, most of these words
are classified as threat-related actions and resources in the context of Israeli cover-
age. Finally, words in Cluster 6 evolve around “opposition” and “election”. Action
words such as “demonstrations” and “protests” reflect the disagreement of the Ira-
nian public with their government and the strive to achieve “democracy”. As this
cluster deals with an ideological and political conflict, it contains both threat- and
opportunity-related words in the eyes of Israeli readers.
In general, it seems that the seven theoretical functions in the conflict (actors,
resources, situation, position, intentions and interests, actions, and results) are spread
among the thematic clusters. Still, there are certain functions that dominate some
of the themes. Thus, for example, Cluster 2 that focuses on missile attacks and
Cluster 4 that focuses on negotiations and peace talks display significantly more
action-related words. A chi-square test confirms the significant correlation between
the word clusters and their functions, χ2(86) = 57.15, p = .002 in Haaretz, and χ2
(81) = 45.35, p = .036 in Yedioth Ahronoth.
Based on the theme—development of nuclear weapons, international con-
flicts, economic sanctions, or peace negotiations—the tone of the words tends
to be more threat- or opportunity-related from the Israeli point of view.
Naturally, there are more threat-related words when it comes to the coverage
and framing of Iran in the Israeli media. Still, certain clusters (such Cluster 4
dealing with peace talks) contain more opportunity-related words, while
others (such as Cluster 2 dealing with missile attacks) contain more threat-
related words. Here too, we found a significant correlation between the word
clusters and their tone, χ2(86) = 29.89, p = .001 in Haaretz, and χ2(81) =
40.67, p < .001 in Yedioth Ahronoth. The relatively high and significant cor-
relation between the tone of words and their network clusters in the latter
confirms that Yedioth Ahronoth tends to display more polarized views. Hence,
as can be seen in Figure 2.4, the clusters dealing with missile attacks and
nuclear developments in Yedioth Ahronoth contain predominantly threat-related
words, while the cluster dealing with peace talks contains mostly opportunity-
related words.
 The news coverage of threats 47

Discussion
In this chapter, we demonstrated the use of semantic network analysis to outline
the threat and opportunities discourse in two leading Israeli newspapers covering
the nuclear developments in Iran over one decade. Our findings reveal similarities
in representing the main narrative told by the leftist newspaper, Haaretz, and the
more centrist newspaper, Yedioth Ahronoth. At the same time, the analysis enables us
to identify several differences between the newspapers, as will be detailed below.
One of the remarkable similarities between the two outlets is that both emphasized
the fear of war between Israel and Iran. In both networks the word “war” was at the
center of the discussion with a very high betweenness. Moreover, the main frames
surrounding the fear of war were very much similar across the newspapers. Both fol-
lowed closely the development of nuclear weapons in Iran, on the one hand, and the
importance of ensuring the security of Israel, on the other. They both described the
negotiations and economic sanctions on Iran as important measures to prevent Iran
from developing nuclear weapons and promoting greater stability in the region.
Finally, the two outlets focused on missile attacks on the border of Israel by Hezbollah
as one of the actual manifestations of the Iranian threat, and described the election of
extremist politicians in Iran as a crucial factor in the escalation of the crisis.
Following the literature, the main reason for the overall agreement between the
two outlets is related to the fact that both rely heavily on official governmental and
military sources. In line with previous studies (Bennett, 1990; Farnsworth et al.,
2010; Livio & Cohen-Yechezkely, 2019; Merom, 2017), the existential threat that
is emphasized by many politicians and their skeptical views on any international
attempt to negotiate with Iran, is reflected in the media. The networks of actors in
both Haaretz and Yedioth Ahronoth support this claim as they both frequently
mentioned official sources in Israel and the US, and hardly any sources from the
opposite side of the conflict. Moreover, the clear distinction between Western
actors and Iranian allies demonstrates well the rhetoric of “us” against “them”,
which is typical to the coverage of international conflicts (Amin, 2019; Izadi &
Saghaye-Biria, 2007; Kadkhodaee & Ghasemi Tari, 2019; Khosravinik, 2015b;
Klein, 2009).
Apart from the reliance on similar sources, it has been already found that news
coverage tends to be supportive particularly when it comes to the question of
national threat (Lavie-Dinur et al., 2018; Randahl, 2018; Yarchi et al., 2015). Since
the course of events greatly depends on the Israeli government and its American
ally, national outlets often adopt the authorities’ point of view and uncritically
support their actions (Feinstein, 2016, 2018; Karniel et al., 2017; Yarchi et al.,
2015; Randahl, 2018).
Together with similarities, our semantic network analysis revealed several
differences between Haaretz and Yedioth Ahronoth. These differences can be
observed when looking at two main aspects: who is mentioned in the story and
how the problem is being addressed. First, in terms of who is being mentioned,
the leftist Haaretz features some Arab states and discusses the Palestinian issue in
48 Hatzir, Segev, Watanabe, and Tago

relation to the Iranian threat, whereas the more centrist Yedioth Ahronoth mentioned
the Arab world much less and completely ignored the Palestinian Authority. This
finding indicates, in accordance with previous studies (Lavie-Dinur et al., 2018), that
Haaretz offers a more complex international and national context to the story than
Yedioth Ahronoth.
Second, in terms of how the problem is addressed, Haaretz focuses on the
international pressure and diplomatic strategies to limit uranium enrichment.
Yedioth Ahronoth, on the other hand, emphasizes the right of Israel to defend itself,
and the importance of concrete actions such as sanctions on Iran to prevent nuclear
development and attacking targets across the border to protect Israel’s security.
These differences reflect the general approach of each newspaper. While the leftist
Haaretz attempts to take a “softer” and more balanced approach, the centrist
Yedioth Ahronoth displays “harder” views on the Iranian nuclear threat. This further
corroborates previous findings on the different styles of the leftist and the more
centrist outlets that emphasize the danger of terror attacks and the right of Israel to
secure its borders (Lavie-Dinur et al., 2018).
To conclude, semantic network analysis was found to be instrumental in identifying
the main narrative of Israeli newspapers covering the Iranian nuclear development—
the threat of war. It also helped to deal with large texts when mapping the main
themes in which Iran is mentioned, such as nuclear developments international
agreements and sanctions, international relations, and the economic interests and
political processes in Iran. Finally, it also revealed the more nuanced differences
between the two outlets, reflecting their different ideologies and political inclinations.
This chapter helps to demonstrate the potential of semantic network analysis, and we
hope it will encourage future studies to further develop and use this method when
exploring media discourse on international conflicts.

Tips and lessons from this process

1. Semantic network analysis is particularly useful when looking at large texts.

In this case, we created networks of words based on all news articles
mentioning “Iran” in two leading newspapers over more than a decade
(between 8000 to 15,000 news articles containing 2.7 to 4.3 million
words, respectively).
2. It is extremely important to choose appropriate words to include in the
network. One way to do so is to carefully inspect the list of most frequent
words and think about how they can be classified into categories that
correspond to the research question. In the case of the coverage of Iran in
Israeli news, we came up with seven functions: actors (e.g., country
names, organizations, and politicians), assessment of the situation (e.g.,
threat, crisis, or war vs. security or peace), stand or position (e.g., extreme
or hostile vs. moderate or friendly), intentions or interests, actions (e.g.,
attack or bombing vs. negotiation or agreement), resources (e.g., weapon
 The news coverage of threats 49

vs. energy), and finally, outcomes or results (e.g., failure vs. success). It is
crucial to go back and forth to the text and understand the context in
which words are mentioned. Many of these words were also related to
threats or opportunities from the Israeli point of view, allowing us to easily
interpret the network according to the theory of media framing during
crisis and threat.
3. Cleaning the network and making it readable is always about optimizing
and maintaining the balance between two extremes—a dense unreadable
network on the one hand, and a sparse simplistic and non-informative
network on the other. As a rule of thumb, it is desirable to achieve a net-
work of about 50 to 100 words that can be classified into between five to
ten distinctive clusters. This would often result in an informative and
interpretable network with meaningful frames and topics.
4. To sparsify the network, one can carefully delete the less frequent ties.
Additionally, when a word in the network appears at the absolute center
and is part of the issue being studied (in our case “Iran” or “Israel”),
deleting this word will often enable a researcher to observe the structure of
the network more clearly. Finally, it might sometimes be useful to create
separate networks for different word functions. Thus, in our case we created
one network for the actors involved in the story (i.e., countries, organizations,
and politicians), and another for the non-actors (i.e., actions, positions, and
intentions). This enabled us to more clearly see the relationships within words
that serve similar functions.
5. When looking for the most central words in each cluster, the measurement
of betweenness, perhaps more than degree centrality, is often instrumental
in semantic network analysis. It creates a clear distinction between central
and less central words, and once betweenness is visualized it helps
researchers to identify the main and most important context around which
other words are discussed in each cluster.

Acknowledgments
This research is part of Grants-in-Aid for Scientific Research (KAKENHI). We
would like to thank the Ministry of Education, Culture, Sports, Science and
Technology (MEXT) in Japan for the generous support. Many thanks also to
Regula Miesch, Alon Hatzir, and Lika Hatzir, who proofread and commented on
earlier drafts of this chapter.

References
Althaus, S. L. (2003). When news norms collide, follow the lead: New evidence for press
independence. Political Communication, 20(4), 381–414.
50 Hatzir, Segev, Watanabe, and Tago

Alarqan, A. (2020). United States position towards Iran after the nuclear deal (2015–2019).
Humanities & Social Sciences Reviews, 8(1), 210–219.
Amin, F. (2019). An “existential threat” or a “past pariah”: Securitisation of Iran and
disagreements among American press. Discourse & Communication, 1–20.
Atai, M. R., & Mozaheb, M. A. (2013). The representation of Iran’s nuclear program in
British newspaper editorials: A critical discourse analytic perspective. International Journal of
Society, Culture & Language, 1(2), 15–33.
Bennett, W. L. (1990). Toward a Theory of Press-State Relations in the United States.
Journal of Communication, 40(2), 103–127.
Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of
communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 10, 10.
Boudana, S., & Segev, E. (2017). The bias of provocation narratives in international news.
The International Journal of Press/Politics, 22(3), 314–332.
Brandes, U., &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In
M. Jünger & P. Mutzel (Eds.), Graph Drawing Software (pp. 321–340). New York:
Springer-Verlag.
Cappella, J. N., Joseph, N., Jamieson, K. H., & Hall, K. (1997). Spiral of Cynicism: The Press
and the Public Good. Oxford: Oxford University Press.
Carter, D. L., Thomas, R. J., & Ross, S. D. (2011). You Are Not a friend: Media Conflict
in Times of Peace. Journalism Studies, 12(4), 456–473.
Danieli, D. (2019). Dangerous nuclear proliferation in the Middle East: An Israeli perspective.
SAIS Review of International Affairs, 39(2), 129–137.
Entman, R. M. (2003). Cascading activation: Contesting the White House’s frame after 9/
11. Political Communication, 20(4), 415–432.
Entman, R. M. (2007). Framing bias: Media in the distribution of power. Journal of
Communication, 57(1), 163–173.
Farnsworth, S., Soroka, S., & Young, L. (2010). The international two-step flow in foreign
news: Canadian and U.S. Television news coverage of U.S. affairs. The International Journal
of Press/Politics, 15(4), 401–419.
Feinstein, Y. (2016). Rallying around the president: when and why do Americans close
ranks behind their presidents during international crisis and war? Social Science History, 40
(2), 305–338.
Feinstein, Y. (2018). One flag, two rallies: Mechanisms of public opinion in Israel during the
2014 Gaza war. Social Science Research, 69, 65–82.
Gamson, W. A., & Modigliani, A. (1989). Media discourse and public opinion on nuclear
power: A constructionist approach. American Journal of Sociology, 95(1), 1–37.
Ghane, M. H., & Mahdavirad, F. (2016). Journalistic stance in newswriting on Iranian
nuclear issue. International Journal of English Linguistics, 6(6), 118–128.
Hayes, D., & Guardino, M. (2010). Whose views made the news? Media coverage and the
march to war in Iraq. Political Communication, 27(1), 59–87.
Hearns-Branaman, J. O. (2017). Official enemies in commercial and soft power media:
Agency and the Iran crisis in the United States, United Kingdom, Chinese, and Iranian
news. Journalism Studies, 18(4), 449–469.
International Atomic Energy Agency (IAEA). (2020). Convention on the physical protection
of nuclear material. Retrieved on February 21, 2020 from https://www.iaea.org/publica
tions/documents/conventions/convention-physical-protection-nuclear-material.
Izadi, F., & Saghaye-Biria, H. (2007). A discourse analysis of elite American newspaper editor-
ials: The case of Iran’s nuclear program. Journal of Communication Inquiry, 31(2), 140–165.
Jaspal, R. (2014). Representing the “Zionist Regime”: Mass communication of Anti-Zionism
in the English-language Iranian press. British Journal of Middle Eastern Studies, 41(3), 287–305.
 The news coverage of threats 51

Jaspal, R. (2015). Antisemitism and anti-Zionism in Iran: The effects of identity, threat, and
political trust. Contemporary Jewry, 35(3), 211–235.
Jones, C. (2018). Israel’s security nexus as strategic restraint: The case of Iran 2009–2013.
Journal of Strategic Studies, 41(1–2), 160–180.
Kadkhodaee, E., & Ghasemi Tari, Z. (2019). Otherising Iran in American political discourse:
case study of a post-JCPOA senate hearing on Iran sanctions. Third World Quarterly, 40(1),
109–128.
Kampf, L., & Liebes, T. (2013). Transforming Media Coverage of Violent Conflicts: The New Face
of War. London: Palgrave Macmillan.
Karniel, Y., Lavie-Dinur, A., & Samuel Azran, T. (2017). Professional or personal framing?
International media coverage of the Israel–Hamas prisoner exchange deal. Media, War &
Conflict, 10(1), 105–124.
Kazemzadeh, M. (2017). Foreign policy decision making in Iran and the nuclear program.
Comparative Strategy, 36(3), 198–214.
KhosraviNik, M. (2015a). Discourse, Identity and Legitimacy: Self and Other in Representations of
Iran’s Nuclear Programme. New York: John Benjamins Publishing Company.
KhosraviNik, M. (2015b). Macro and micro legitimation in discourse on Iran’s nuclear
programme: The case of Iranian national newspaper Kayhan. Discourse & Society, 26(1),
52–73.
Klein, A. (2009). Characterizing “the Enemy”: Zionism and Islamism in the Iranian and
Israeli press. Communication, Culture & Critique, 2(3), 387–406.
Lavie-Dinur, A., Yarchi, M., & Karniel, Y. (2018). The portrayal of lone wolf terror wave
in Israel: an unbiased narrative or agenda driven? The Journal of International Communication,
24(2), 196–215.
Liebes, T. (1997). Reporting the Arab-Israeli conflict: How Hegemony Works. New York:
Routledge.
Liebes, T., & Kampf, Z. (2009). Black and white and shades of gray: Palestinians in the Israeli
media during the 2nd Intifada. The International Journal of Press/Politics, 14(4), 434–453.
Livio, O., & Cohen-Yechezkely, S. (2019). Copy, Edit, Paste: Comparing news coverage of
war with official military accounts. Journalism Studies, 20(5), 696–713.
MacDonald, M. N., Homolar, A., Rethel, L., Schnurr, S., & Vessey, R. (2015). Manu-
facturing dissent: The discursive formation of nuclear proliferation (2006–2012). Discourse
& Communication, 9(2), 173–197.
MacDonald, M. N., & Hunter, D. (2019). Discourse of nuclear proliferation. In The Dis-
course of Security: Language, Illiberalism and Governmentality. New York: Springer.
Maher, N. (2020). Balancing deterrence: Iran-Israel relations in a turbulent Middle East.
Review of Economics and Political Science.
Merom, G. (2017). Israeli perceptions of the Iranian nuclear threat. Political Science Quarterly,
132(1), 87–119.
Mousavian, S. H., & Mahmoudieh, Y. (2017). The nuclear deal: A crossroad or deadlock in
relations with Iran. International Studies Journal, 14, 143–168.
Mousavian, S. H., & Mousavian, M. M. (2018). Building on the Iran nuclear deal for
international peace and security. Journal for Peace and Nuclear Disarmament, 1(1), 169–192.
Neiger, M., Zandberg, E., & Meyers, O. (2010). Communicating critique: Toward a con-
ceptualization of journalistic criticism. Communication, Culture & Critique, 3(3), 377–395.
Nossek, H. (2004). Our news and their news: The role of national identity in the coverage
of foreign news. Journalism, 5, 343–363.
Parsi, T. (2013). A Single Roll of the Dice: Obama’s Diplomacy with Iran. New Haven, CT:
Yale University Press.
52 Hatzir, Segev, Watanabe, and Tago

Pedraza, J. M. (2017). Are nuclear-weapon states not parties to the non-proliferation treaty
ready to renounce to the possession of nuclear weapons? Public Organization Review, 17(3),
335–352.
Pieper, M. (2019). An Iran nuclear deal without the United States? Chinese, European, and
Russian interests and options after the US Withdrawal from the Joint Comprehensive
Plan of Action. In Contemporary Issues in International Political Economy (pp. 35–53). Sin-
gapore: Palgrave Macmillan.
Randahl, D. (2018). Terrorism and public opinion: The effects of terrorist attacks on the popu-
larity of the president of the United States. Terrorism and Political Violence, 30(3), 373–383.
Seybert, A. (2019). Nuclear politics in the Mediterranean: Prospects and pitfalls of nuclear
weapons programs in Middle Eastern states. Mediterranean Politics, 24(3), 384–391.
Segev, E. (2016). International News Online: Global Views with Local Perspectives. New York:
Peter Lang.
Segev, E. (2020). Hebrew stop-words. https://github.com/koheiw/marimo/blob/master/
yaml/stopwords_he.yml.
Slovic, P. (2004). The Perception of Risk. London: Earthscan.
United Nations (UN). (2020). Treaty on the Non-Proliferation of Nuclear Weapons
(NPT). Office for Disarmament Affairs. Retrieved on February 26, 2020 from: https://
www.un.org/disarmament/wmd/nuclear/npt/.
Von Sikorski, C., Schmuck, D., Matthes, J., & Binder, A. (2017). “Muslims are not terror-
ists”: Islamic State coverage, journalistic differentiation between terrorism and Islam, fear
reactions, and attitudes toward Muslims. Mass Communication and Society, 20(6), 825–848.
Woods, J. (2011). Framing terror: An experimental framing effects study of the perceived
threat of terrorism. Critical Studies on Terrorism, 4(2), 199–217.
Yarchi, M., Galily, Y., & Tamir, I. (2015). Rallying or criticizing? Media coverage of
authorities’ reaction to terror attacks targeting sporting events. Studies in Conflict & Ter-
rorism, 38(12), 1008–1021.
Yazdani, E. (2019). Iranian foreign policy during Rouhani Presidency: Perspective on
change and continuity. International Relations and Diplomacy, 7(10), 472–484.
3
PROVOCATION NARRATIVES IN
CHINESE AND US NEWSPAPERS
Yuan Zhou, Sandrine Boudana and Elad Segev

Introduction
Research shows that news media tend to align their coverage of foreign affairs with
their national government’s positions. This means adopting frames that resonate
with the official geopolitical vision and, in case of armed conflicts, attributing
responsibility for the violence accordingly.
As Boudana and Segev (2017a) have demonstrated, provocation narratives
represent one way in which news practitioners may shift blame to the victims
whenever the identity of perpetrators of violent deeds does not match acceptable
storylines. Their examination of news articles containing the word “provocation”
in Western newspapers revealed that North Korea is the most frequently cited
country in those articles and is mostly framed as a provoking country, whereas the
noticeable absence of provoked countries in many of these articles can be inter-
preted as imposing a universal, rather than partisan, perspective. North Korea is
implicitly designated not as the adversary of South Korea and the US, but as the
enemy of all Western countries, if not the whole world.
The story told by Western media about the threat that North Korea poses to the
stability of the region and the entire world is certainly one-sided. As expected as it
may be for the failure of news media in countries supporting North Korea to
provide impartial accounts on the conflict. But how does the reversal of the roles
of victims and perpetrators operate in those supporting countries? Can we observe
symmetrically opposite narratives of provocation from the news media in countries
that support and oppose North Korea? In this chapter, we complete and update the
analyses of the use of provocation narratives in US news coverage of North Korea,
and compare it with the story told by the Chinese press. Given the growing ten-
sion between the US and China, we would expect that the Chinese media would
frame the Korean conflict differently. We conclude on the implications of our

DOI: 10.4324/9781003120100-4
54 Y. Zhou, S. Boudana, and E. Segev

study for the research on foreign news, news frames, press-state relations, and for
the notion and limitations of press freedom.

News media and foreign policy

Since Cohen’s (1963) classic book, scholarly discussions on the relations between
news media and foreign policy have been wide, intense, and not exempt from
contradictory claims (Malek, 1997). Some scholars have looked at the impact that
reporters and editors have on their government’s foreign policy, notably by shaping
public opinions (Herman, 1993; Powlick & Katz, 1998; Robinson, 1999; Robin-
son, 2002). Others have examined the various ways in which national governments
influence the media coverage of foreign news, essentially based on two theories:
hegemony and indexing (Entman, 2003).1 According to the first theory, news
practitioners more or less consciously align their coverage with the national elites’
positions (Altschull, 1984; Hallin, 1987; Lee & Yang, 1995; Zaller & Chiu, 2000;
Handley & Rutigliano, 2012), even more than regarding domestic affairs (Gans,
1979; Graber, 1989).
The second theory considers that the media have more leeway in covering foreign
affairs than implied by a propagandistic model, as they reflect the unavoidable dis-
agreements among political elites (Bennett, 1990). However, in that latter case too,
the freedom and power that news media seem to entertain are constrained by the
frames that the political elites have predefined (Hackett, 1984). Thus, even when
official sources and opposing views are equally represented by the media, the former
remains the primary definers of the reported event, whereas the latter’s role is
restricted to reacting and contradicting them within the boundaries drawn by the
dominant frame (Sigal, 1973; Hall et al., 1978; Bennett, 1996).

Frames and blame games in foreign news

In his seminal paper clarifying the concept of news frames, Entman (1993) warns
that “because they lack a common understanding of framing, journalists frequently
allow the most skillful media manipulators to impose their dominant frames on the
news” (pp. 56–57). Thus, Entman proposes a practical definition of framing, based
on the identification of its four functions: “problem definition, causal interpretation,
moral evaluation, and/or treatment recommendation” (p. 52). The way these func-
tions are assigned or not assigned to specific actors is likely to influence how audiences
will perceive responsibilities for, and possible solutions to, a problem—and even how
they will define the problem in the first place.
Attribution of responsibility to specific actors may help to solve problems. For
instance, McClure et al. (2010) have shown that the fatalistic coverage of natural
disasters tends to convince audiences that damages cannot be prevented, whereas
causal information attributing responsibility to political and other actors would
facilitate preventive actions. However, in many other cases, agency attribution leads
to a counter-productive blame game (Knobloch-Westerwick & Taylor, 2008). For
 Provocation narratives 55

instance, Iyengar (1990) argues that when frames about poverty focus on indivi-
duals, responsibility for poverty is attributed to specific actors (the poor people),
rather than seen as a social issue calling for policy change.
In the coverage of foreign conflicts, attribution of responsibility is more akin to
this sort of blame game. Indeed, foreign actors involved in a conflict and their
supporters among national elites are mobilized to push narratives that shift the
responsibility for causing or fueling the conflict to the other side (Boudana, 2010).
Such pressure accentuates the trends toward polarization, oversimplification, and
“emotionalization” in news reports (Peters, 2011). Finally, journalists may more
easily propose unsophisticated polarized narratives on foreign conflicts because their
audiences, and sometimes themselves, lack knowledge about these conflicts, their
historical background, and geopolitical implications (Welch, 1972; Paletz &
Entman, 1981). Therefore, foreign conflicts are often framed in terms of blaming
the “bad guys” (Campbell, 1972; Barkin, 1984; Boudana, 2015; Segev, 2016), a
role often cast in accordance with the national government’s position.

Provocation narratives in the Korean conflict

In order to maintain the assignment of the “bad guy” role to the same actors in
a conflict, narrators may sometimes have to shift blame to the victims of vio-
lence. Boudana and Segev (2017b) have identified provocation narratives as an
efficient “communication strategy” to serve that purpose (p. 2). Violence is
then presented as a natural, expected, and therefore legitimate reaction to pre-
ceding words or actions characterized as blameworthy. By deleting or reversing
causality, provocation narratives thus pass responsibility for the violence to the
victims. Feminists and gender scholars (Meyers, 1994; Meyer, 2010; Tilleman,
2010) have been among the first to sensitize the public to the bias inherent in
such narratives, which are notably found in news reports on rapes and murders
of female victims.
Theorizing and measuring such bias in international news, Boudana and
Segev (2017a) have identified the Korean conflict as the most prominent in
provocation narratives used by Western news media (American, British,
German, and French news). Using semantic network analysis of countries co-
mentioned together in the same news item, they show that provocation narra-
tives in the international news of Western countries almost always dealt with
the “bad guys” or the enemies of the US and its Western allies. In these nar-
ratives North Korea is by far the most prominent provocateur, often depicted
as an aggressive country, launching long-term missiles, and developing nuclear
weapons. Yet there is no attempt to explain or justify its actions or motives
beyond the notion of “provocation”. Boudana and Segev’s (2017a) findings
corroborate other studies on elite American newspapers displaying an anti-
North Korea frame in their news coverage (Choi, 2010; Kim, 2014; Seo &
Nah, 2020) in line with the US government official views.
56 Y. Zhou, S. Boudana, and E. Segev

Extending to a comparison with Chinese newspapers

In this chapter, we update Boudana and Segev’s (2017a) study on US newspapers
and extend it to a comparison with Chinese newspapers. In their review of the
communication research on North Korea, Seo and Nah (2020) identified South
Korea as the most important site of research, followed by the US, whose scholars
often drew comparative analyses, notably with South Korea. Then came Japan and
China, which are neighboring countries that participate in international negotiations
to solve the conflict, such as the six-party talks. The US and China thus represent
two of the most covered countries in academic research on North Korea, as they are
major players in the Korean conflict over the past decades. They are regarded as
supporters of South and North Korea, respectively. In our study, we anticipate that
the American and Chinese news media will align with those official positions.
According to Seo and Nah (2020), scholars concur in their findings that the US
media, and the Western media more broadly, depict the Korean conflict in sim-
plistic terms, and paint a dehumanizing portrait of North Korea. As for the Chinese
media, Hinck and Cooley (2020) found that they tended to embrace North
Korean narratives when North Korea’s actions aligned with the Chinese govern-
ment’s interests. This translated into news coverage that “included legitimizing the
Kim regime, bolstering Russian and Chinese international influence, and reducing
US influence and support for denuclearization” (p. 1331). Yet questions of geo-
graphic proximity and recent evolutions in the geopolitical relations between
North Korea and the American and Chinese governments could explain why the
news coverage of the Korean conflict might not be so clear-cut and symmetrically
opposed in the US and Chinese newspapers.
First, in contrast with Hinck and Cooley’s findings, Seo and Nah (2020) point out
that in China and Russia, which support North Korea, the press was found to “over-
whelmingly adopt a peace journalism frame, focusing more on nonviolent compromises
and voices of the common people” (Chung, Lessman, & Fan, 2014, p. 1316). Scholars
explain this difference by the fact that Chinese and Russian news outlets are more easily
granted access to North Korea and can more directly and extensively report from the
field. This is in line with Segev’s (2016) finding that geographic proximity leads to more
nuanced and detailed news coverage. Thus, the Western Manichean depiction of the
conflict contrasts even with the coverage in South Korea, which is “less negative and
more nuanced” as “South Korean media is more likely to use peace frames and
emphasize reconciliation on the Korean peninsula” (Seo & Nah, 2020: 1316).
Furthermore, China has more complex diplomatic cards to play and, unlike the US,
can hardly adopt a Manichean position based on a universalist perspective. Indeed,
Boudana, and Segev (2017a) found that provocation stories in the American media
often omitted to mention the provoked countries (i.e., South Korea and the US), thus
implying that the international community, rather than specific entities, is threatened
by North Korea. But China does not have the same latitude to present its ally’s ene-
mies (i.e., South Korea) as the enemies of the world, because it is a permanent
member of the UN, which generally condemns North Korea’s military actions.
 Provocation narratives 57

Finally, recent evolutions of the diplomatic relations between North Korea and
the Chinese and American governments may account for less clear-cut reporting.
In particular, scholars (Zhan, 2016; Zhang & Zinoviev, 2018) have pointed out
that China’s support for North Korea has waned in the past years. Under interna-
tional pressure, China significantly reduced its exports of food, fuel, and oil
between 2012 and 2015, and has taken more drastic measures since 2017, including
embargos on the import of some basic goods. Those measures have not deterred
North Korea from carrying out more missile tests. Meanwhile, the relations
between North Korea and the US have warmed up, culminating in the historic
encounter between the two countries’ leaders in June 2019. Boudana and Segev’s
(2017a) study of US news was based on a sample of items collected between 2013
and 2016. The news coverage between 2016 and 2019 presented in this chapter
may be more nuanced.
The three factors mentioned—the question of geographic proximity, China’s
complex geopolitical game, and the recent changes in the diplomatic relations
between the actors involved in the conflict—could explain the possible differences
between the American and Chinese news coverage. However, since we are only
interested in news items containing the word “provocation”, it is likely that such
items use more polarized narratives than other news stories. Therefore, we can still
hypothesize that provocation narratives found in the US and Chinese newspapers
will present a reverse polarization: in US newspapers, North Korea will be the
provoking country, South Korea the provoked countries, and the US the accusing
entity; whereas, in Chinese newspapers, North Korea will be the provoked coun-
try and South Korea the provoking country, with China playing the role of the
accusing entity.
The comparison between Chinese and US newspapers will also be instructive
regarding the journalism cultures in those countries. In American newspapers, jour-
nalists rarely take the role of the accusing party. This is not so much because of an
actual critical distance from the provocation frame (since most concur and put North
Korea in the role of the accused, in line with their national government), as because
the standard of objectivity requires the conventions of detachment and neutrality to
be respected. Thus, if and when Chinese journalists endorse their government’s
positions, we expect this to be done in a more straightforward manner.

Methods

Sample
We collected all articles that mentioned the word “provocation” in English from
two American newspapers, the New York Times and the Washington Post, and all
articles that mentioned the word “挑衅” in Chinese from two Chinese news-
papers, People’s Daily and Global Times, over a period of three years between
November 2016 and October 2019.2 We chose those four newspapers because of
their popularity and representativeness in their respective countries.
58 Y. Zhou, S. Boudana, and E. Segev

The New York Times and the Washington Post are among the most influential
elite newspapers in the US. The more liberal views of the New York Times and the
more conservative leaning of the Washington Post may be reflected in different
depictions of the actors in reporting on the Korean conflict. In China, People’s
Daily is one of the most authoritative official media. It is published and directly
controlled by the Central Committee of the Communist Party of China, and its
coverage of political topics is considered to reflect the views of Chinese leadership
(Wu, 1994). Global Times is a daily tabloid, which is famous for its nationalistic
perspective and pro-government tone. The Chinese government has enacted a
series of laws and regulations that prohibit non-official media and foreign media
from producing political news in China.3 Global Times is one of the few commer-
cial media that are allowed by the government to introduce news content from
foreign media to the Chinese domestic audience. It selectively translates and adds
comments to news from mainstream media of other countries (most are negative
news about foreign countries, and some are international praise of China). Com-
pared with People’s Daily, the language of Global Times is more straightforward and
radical. It has considerable influence on Chinese public opinion.
We focus on the period of November 2016 to October 2019, which, as we
pointed out, has been marked by significant geopolitical changes in the Korean
conflict. The semantic network analysis included two main stages. The first stage
consisted of mapping and comparing the countries related to provocation narratives
as presented in US and Chinese news. For this purpose, we constructed networks
of country names mentioned together in the same news item. To detect the
countries mentioned in provocation news, we created a dictionary composed of
the names of the 193 UN member countries both in English and Chinese.4 For
each country name, we calculated the number of articles containing that name in
each newspaper.
The second stage consisted of comparing the provocation narratives based on the
frequent words appearing together in the same sentence. We followed the procedure
discussed in earlier chapters, including identifying the frequent words in English and
Chinese news, removing stop-words and words that did not relate to the provoca-
tion narratives, identifying the central words and prevalent combinations of words,
and finally analyzing the clusters of words that appeared more frequently in each
network. Once we identified the clusters in the network, we tried to better under-
stand the connections between the words by looking at the contexts of their use in
randomly selected articles.

Results

Prominent countries in provocation context

Figure 3.1 shows the monthly number of articles of each newspaper in our analysis.
What stands out in the figure is that Global Times uses provocation narratives much
more than the other three outlets. This result confirms that Global Times upholds
 Provocation narratives 59

FIGURE 3.1 Monthly number of provocation articles in each newspaper.

the banner of nationalism and frequently uses sensational rhetoric like provocation
as part of its competition strategy in the Chinese media market (Lu, 2013).
Table 3.1 shows the top ten countries mentioned in relation to provocation in
each newspaper. It appears that in all four newspapers provocation stories are
primarily related to the US, China, Russia, Japan, North Korea, and South
Korea. For each of these top mentioned countries we further investigated the
ratio between the number of news articles that mentioned them together with
the word “provocation” and the total number of articles mentioning them
(Figure 3.2).
As can be seen, North Korea has the highest ratio in both the US and China,
indicating that when North Korea is mentioned it is more likely to be mentioned
in the context of provocation than any other country. In general the figures are
much higher in Chinese than in US newspapers. One possible reason is that we
only retrieved articles containing the word “provocation” from US newspapers,
while the Chinese equivalent “挑衅” can be both noun (provocation) and verb
(provoke). If we added articles containing “provoke” and its variations, the per-
centage in US newspapers would increase, but the context of articles would
become less relevant (as the word “provoke” in English can appear in various
contexts, for example, “provoke a discussion”). Another reason is that Global Times
is notorious for its nationalist narratives and thus has a greater tendency to use
sensationalistic frames, including provocations (see Table 3.1).
60 Y. Zhou, S. Boudana, and E. Segev

TABLE 3.1 Top countries mentioned in provocation related news

All news New York Times Washington Post Global Times People’s Daily
Country % Country % Country % Country % Country %
US 74.8 US 60.6 US 81.4 US 81.9 China 79.7
China 54.6 China 24.2 Russia 37.1 China 72.8 US 65.0
Russia 30.3 Russia 23.1 North 27.4 Japan 36.6 Russia 23.1
Korea
Japan 28.8 France 19.6 China 26.6 Russia 33.3 Japan 21.0
North 26.5 Japan 18.9 South 22.8 North 33.2 Germany 18.9
Korea Korea Korea
South 23.6 North 17.6 Japan 20.3 UK 31.5 UK 15.4
Korea Korea
UK 20.2 Syria 13.4 Syria 13.5 South 30.9 South 11.2
Korea Korea
Germany 17.4 Iran 13.0 Iran 12.7 Germany 21.1 France 11.2
France 14.3 South 12.8 Israel 11.8 France 13.0 Iran 10.5
Korea
India 10.7 Germany 12.7 Iraq 11.0 Australia 11.7 Syria 9.1
N 2006 545 237 1081 143

FIGURE 3.2 The ratio of provocation related news per country in Chinese and US
media.
 Provocation narratives 61

Networks of countries in provocation news

Figure 3.3 represents the network of countries mentioned in US news in relation
to provocation. As can be seen, the most central countries related to provocation
narratives in US news are the US itself, Russia, Iran, and China (visualized based
on nodes’ Betweenness centrality).
A cluster analysis based on Louvain modularity reveals six clusters representing
main regional conflicts: The Korean conflict (bottom-right, including North and
South Korea as well as the US, China, and Japan), the Russian conflict (bottom-
left, including Russia, Ukraine, and other neighboring countries), and the Middle
Eastern conflict (top-left, including Iran, Israel, and other neighboring countries).
Other smaller clusters include France and other European countries (top-right,
with connection to both the Korean and the Middle Eastern conflicts), as well as
more specific regional conflicts such as Venezuela and Columbia (bottom-right),
and Pakistan and Afghanistan (bottom-center).

FIGURE 3.3 Undirected network of country co-occurrences in US news.

Note. The size of the nodes represents their Betweenness centrality, the width of the ties
represents the number of news items in which each pair of countries are mentioned.
Five top links are marked in dark gray. For visualization purposes, we include only
countries that appeared in 100 news items or more. Clusters are based on Louvain
modularity. Network visualization and calculations were performed with Visone
(Brandes & Wagner, 2004).
62 Y. Zhou, S. Boudana, and E. Segev

France is prominent in these smaller clusters as it is a major player of the European

Union, but also because, since the Cold War, it often took diplomatic positions that
diverged from the US (Wall, 2008; Bozo, 2019). This includes supporting the Palesti-
nians and more critical position regarding Israel, refusing to join the US-led coalition in
the 2003 Iraq war, and supporting the negotiations with Iran on the nuclear program.
Figure 3.4 displays the network of countries mentioned together in same news item
in Chinese news. As in US news, the most central countries in Chinese news are the
US, China, and Russia. Yet there are apparent differences between the two networks,
reflecting the different focus in the two languages. First, the gap in terms of country
centrality is wider in Chinese news. There is certainly much more news about North
Korea, South Korea, and Japan (as showed by the width and color of the links). Still,
the centrality of North Korea is much lower, and as can be seen from the cluster
analysis, there are many more non-regional actors involved in the Korean conflict,
including Germany and the UK (bottom-left). Russia is mentioned in a separate
cluster but again not in relation to regional conflicts as much as in the broader inter-
national context, including Turkey, Iran, and Yemen (bottom-right). Naturally,
China itself is mentioned in a much broader international context, notably in relation
to many African countries, such as Ethiopia, Gambia, and Somalia. Finally, apart from
the very prominent triangular US–China–Russia relations, the US is mentioned in
relation to American countries, but also African and Middle Eastern countries.

FIGURE 3.4 Undirected network of country co-occurrences in Chinese news.

Note. Network visualization and calculations were performed with Visone (Brandes &
Wagner, 2004).
 Provocation narratives 63

In short, it seems that while the US news portrays distinctively regional conflicts
and attempts to give each one a stage, Chinese news focuses predominantly on
China and its relationships or tensions with the US and Russia, occasionally giving
a broader overview of its international rather than regional impact.
The next step is to look more closely at the frames related to the provocation
narratives in the US and China. For this purpose, we conducted a semantic net-
work analysis. The English corpus contained 1,075,289 words, while the Chinese
corpus contained 788,263 words. First, we identified the top 100 frequent words
related to provocation narratives. We omitted stop-words (such as “can” and
“just”) and words with multiple meanings (such as “time” and “part”). The pre-
processing of Chinese text is mostly the same as in English. One major difference
was the need to conduct word segmentation (tokenization) since there are no
spaces between words in Chinese. We used both the quanteda package (Benoit et
al., 2018) and the jiebaR package (Qin & Wu, 2019) to tokenize the Chinese text,
and the jiebaR package produced better segmentation based on our qualitative
validation. Then, for each corpus, we obtained a matrix of frequent words co-
occurring in the same sentence. Figure 3.5 presents the semantic network of the
US corpus.

FIGURE 3.5 Semantic network of frequent words in US news.

Note. The size of the nodes represents their Betweenness centrality, the width of the ties
represents the number of news items in which each pair of countries are mentioned.
For visualization purposes, we include only countries that appeared in 100 news items
or more. Clusters are based on Louvain modularity. Network visualization and calcu-
lations were performed with Visone (Brandes & Wagner, 2004).
64 Y. Zhou, S. Boudana, and E. Segev

As can be seen in Figure 3.5, there are five clusters representing five different
themes related to provocation narratives. The largest cluster is around the word
“trump” (on the right), referring to the US President, who has attracted intense media
attention due to his unusual speech style interpreted as provocative. Words in this
cluster include “speech”, “policy”, “Twitter”, “election”, and “campaign”, referring
to his major communication channels and activities. In the middle, there is a second
cluster centered around the US, with words such as “officials”, “White House” as well
as “Russia” and “Iran”. If the previous cluster represented the personal aspect of the
US President, this cluster represents the national aspect and the international tensions
between the US and Russia and Iran. The third and relatively large cluster is related to
the most important provocation story in American media of the North Korean nuclear
plan. It includes words such as “military”, “force”, “weapon”, “nuclear”, “missile”,
and “security”. Two smaller clusters dealt with more specific themes. One semantic
cluster emerged around “black” and “people” (at the bottom), representing national
and international protests and campaigns for race equality. Another cluster of words
appeared around the notion of “art” with the words: “museum”, “artist”, and “exhi-
bition” (on the left), representing a very different type of provocation narratives.
Figure 3.6 portrays the semantic network of provocation narratives in China.
Unlike the American news that includes a mixture of personal, national, and
international stories from the political and cultural spheres, provocation narratives

FIGURE 3.6 Semantic network of frequent words in Chinese news.

Note. Network visualization and calculations were performed with Visone (Brandes &
Wagner, 2004).
 Provocation narratives 65

in China evolve solely around international news. It clearly presents the centrality
of three countries: China, the US, and North Korea. Interestingly, the very strong
dyads US–China and US–North Korea suggest that the US rather than China is
represented as the most central country in Chinese provocation narratives. This is
because, in the narratives of Chinese media, the US is usually depicted as a perpe-
trator taking provocative actions all over the world. The cluster of words sur-
rounding China (on the right) includes relatively positive words such as
“important”, “constructive”, “power”, “peace”, “diplomatic”, suggesting that
China portrayed as an important international player in maintaining stability in the
region. This power is mainly achieved through its trade relationships with the US
(the cluster on the left), containing the words “china-us”, “trading” as well as
“hope” and “support”. Like the other most mentioned words, “support” here
refers to the promotion of peace in the region, rather than partisan support that
would polarize the narrative.
The North Korean threat is indeed an important part of the narrative. Here it
seems that China partially adapted the Western narrative as the words surrounding
North Korea include “threat”, “attack”, “sanctions”, and “provocative”. However,
looking at the context in which these words appear in the text, it seems that
Chinese media tell the story from a different perspective. For instance, an article
reporting the US–South Korea military exercises cites the North Korean news-
paper Rodong Sinmun that “the US and South Korean war hawks must be mindful
that the more you provoke, the guiltier you are. The disastrous gunpowder of
revenge will be packed, and the devastating moment will be approaching”. Also, it
quotes the words of the vice-chairman of the Russian State Duma Committee on
Defense that “the US–South Korea military exercises near the border of North
Korea is totally a provocative action” (Yanxi, 2017, author’s translation). But
unlike the dichotomous Western narrative, which presents sanctions as the only
way to control the nuclear engagement of North Korea, the Chinese narrative also
includes softer diplomatic measures. A small cluster of words between China and
North Korea (at the top) includes words such as “cooperation”, “international”,
and “relationship”.
Two smaller clusters (at the bottom) refer to international regions in which China
has interests. One cluster contains words such as “interest”, “economic”, “strategic”,
and “development”. It suggests that rather than employing military coercion, China
tends to use economic statecraft to achieve its political and strategic goals. Another
cluster puts “Taiwan” at the center surrounded by words such as “solve”, “problem”,
“South China Sea”, and “cross-strait”. This cluster reflects China’s concern about the
threats to its territory and sovereignty. Specifically, Chinese media framed the foreign
interference in the South China Sea and Taiwan as provocative actions.
In short, the US provocation narratives are thematically diverse as they refer to
personal aspects (focusing on Trump), international relationships between the US
and other countries, in particular North Korea, clearly defined as provocative, and
finally more specific types of provocations around Black protests and the field of
arts. When it comes to the international scene, however, it is more univocal,
66 Y. Zhou, S. Boudana, and E. Segev

positioning North Korea as the provocative side of the story. In contrast, the pro-
vocation narratives in Chinese news focus predominantly on Chinese diplomatic
relations and offer a more balanced perspective. They emphasize China’s role as a
mediator to promote a peaceful solution between the US, a foreign provocateur, and
North Korea. Our semantic network analysis shows that Chinese news refers to the
Western narrative on the North Korean conflict but adds a critical perspective to it.

Discussion
Provocation narratives represent one strategy in which news practitioners frame the
“bad guys” (Segev, 2016) and assign blame. They can be used to shift responsibility for
violent events to the victims, especially when the identity of perpetrators of the deeds
does not match acceptable storylines. In their prior research, Boudana and Segev
(2017a) have demonstrated how North Korea was the most frequently cited country
in provocation narratives from the US mainstream news media, and how such narra-
tives introduced bias in news stories by dispensing the narrators from providing
explanations to North Korea’s actions: those actions were framed as pure provocations,
while the noticeable absence of provoked countries in many of these articles further
stigmatized North Korea as representing a threat to the whole world.
This study aimed to update those findings and extend the analysis to a
comparison with the Chinese news media, asking whether the latter would
propose symmetrically opposite narratives of provocation as the Chinese gov-
ernment is, historically, a supporter of North Korea. It was therefore expected
that the Chinese media would focus on the US military presence in the Korean
peninsula as a provocation on North Korea, and would emphasize the US
responsibility in the escalation of the conflict and the tension.
Findings on provocation narratives in our sample of US newspapers, articles
published during 2016–2019, are very similar to the conclusions drawn from
Boudana and Segev’s (2017a) research on years 2013–2016. They corroborate
other studies on elite American newspapers displaying an anti-North Korea frame
in their news coverage (Choi, 2010; Kim, 2014; Seo & Nah, 2020) in line with US
government official views. Thus, the warming up of the relationships between the
US and North Korean leaders since mid-2019 had not (yet) been translated into
the news coverage, at least as far as provocation narratives are concerned. Nor did
we find significant differences between the New York Times and the Washington
Post, the more liberal and more conservative US newspapers, respectively.
Noteworthy differences were found between the US and Chinese provocation
news stories. First, while US newspapers referred to provocations in a broader
range of contexts, including arts, provocation narratives in Chinese newspapers
dealt almost exclusively with international relations. They were also more cen-
tered on China as a key actor in diplomatic relations, challenging the US role.
However, despite such privileged attention given to their own country, the
Chinese newspapers showed more nuances in casting roles of provoking and
provoked countries.
 Provocation narratives 67

They adopted, to some extent, the US rhetoric, and the stories were very much
focused on North Korea as the main provocateur. Yet the assignment of responsi-
bility for the provocative acts, and potential reactions to them, shifted. While in
American news the responsibility is clearly assigned to North Korea, in Chinese news
the responsibility is ultimately attributed to the US, depicted as the stronger power,
in a position to react in a more responsible way to resolve the tension. Chinese
newspapers thus emphasized the competition between the US and China for the role
of mediator in the conflict.
Differences were also apparent between the two Chinese newspapers included in
our sample. The Global Times proved to be more straightforward and radical than
the People’s Daily. The nationalistic positioning of the Global Times may be an
important explanatory factor, but so could be the commercial strategy of the
newspaper, which involves more sensationalism in the tone, writing style, and
types of facts reported.
Thus, contrary to our expectations, we did not find provocation narratives to be
symmetrically opposed in the US and Chinese newspapers. Provocation narratives
in Chinese newspapers were not as Manichean as in the US press. Changes in
recent years in the diplomatic relations between the actors involved in the conflict
and China’s complex geopolitical game are important factors to consider. It is also
likely that proximity, and first geographic proximity, between China and North
Korea, played a key role in this nuanced coverage. This is in line with Segev’s
(2016) emphasis on the importance of geographic, political, and economic proxi-
mity in reporting on foreign countries. In the particular case of North Korea, Seo
and Nah (2020) have also pointed out the importance of proximity, notably
expressed by the fact that North Korea has granted Chinese reporters direct access
to the field, access that has been denied to Western journalists.
Certainly, the results of our comparative analysis are surprising if we consider
that the US press entertains more freedom than Chinese newspapers and thus had
more leeway to propose nuances and diversity in their assignment of praise and
blame to the different actors of the conflict and to deviate from the government’s
official line. Yet it appears that geopolitical complexities and (geographic) proxi-
mity between the reported and reporting countries may have played a more
important role than press freedom in the capacity of news media to offer a nuanced
representation of an international conflict.

Tips and lessons for the use of semantic network analysis

1. Semantic network analysis is a useful method that allows us to detect the

collocations of words. In doing so, we can find systematic patterns of how a
story is framed. By applying semantic network analysis to the news articles
mentioning “provocation” in Chinese and US newspapers, we revealed the
different narrative strategies using by media in the two countries.
68 Y. Zhou, S. Boudana, and E. Segev

2. Researchers can choose which words to include in the network analysis

based on their research purposes. In this article, we first conducted net-
work analysis only on the country names to map the main international
actors in provocation news. Then, we analyzed the semantic network of the
top 100 words to further examine how the provocation narratives in Chinese
and US newspapers diverge from each other.
3. The pre-processing of some languages may be different than others. In
this study, we use the jiebaR algorithm (Qin & Wu, 2019) to make the
Chinese text segmentation more accurate. Once we have created a
matrix of co-occurring words, semantic network analysis can proceed
regardless of the language.
4. Automatic content analysis itself cannot tell the whole story. To validate and
interpret the results of the semantic network analysis, researchers should
scrutiny the original documents related to the key findings. In our case, we
carefully read some articles in both Chinese and US newspapers and raised
some typical examples of provocation narratives. This helped us to “zoom in”
on some central words in the network and contextualize our findings.

Notes
1 Entman (2003) proposes a “cascading activation model”, meant to “supplement(s)
research using the hegemony or indexing approaches”: The model explains how interpretive
frames activate and spread from the top level of a stratified system (the White House) to the network
of non-administration elites, and on to news organizations, their texts, and the public—and how
interpretations feed back from lower to higher levels (p. 415).
2 Articles of the New York Times and Washington Post were downloaded from the Lexis Advance
database, and articles of People’s Daily and Global Times from the WiseNews database.
3 For example, the Provisions for the Administration of Internet News Information Ser-
vices (2017) and the Catalogue of Industries for Guiding Foreign Investment (2004).
4 We use regular expressions to reduce ambiguity when matching country names. For
example, Haiti (海地) can match coastal area (沿海地区/沿海地带) in Chinses, so we use
“海地(?![带区])” to exclude the mismatch. Another example in English is that “Congo”
can match both Congo and Democratic Republic of Congo. In this case, we use “(?<!
Democratic Republic of)Congo” to match the former, and “Democratic Republic of
Congo” to match the latter.

References
Altschull, H. J. (1984) Agents of Power: The Role of the News Media in Human Affairs. White
Plains, NY: Longman.
Barkin, Steve M. (1984) The journalist as storyteller: An interdisciplinary perspective.
American Journalism 1(2), 27–33.
Bennett, W. Lance (1990) Toward a theory of press-state relations in the United States.
Journal of Communication 40(2), 103–125.
Bennett W.Lance (1996) An introduction to journalism norms and representations of politics.
Political Communication 13(4), 373–384.
 Provocation narratives 69

Benoit, Kenneth, Watanabe, Kohei, Wang, Haiyan (2018). quanteda: An R package for the
quantitative analysis of textual data. Journal of Open Source Software 3(30), 774.
Bozo, Frederic (2019) La politique étrangère de la France. Paris: Flammarion.
Boudana, Sandrine (2010) Journalistic objectivity as a performance. The case of the French press coverage
of the Second Intifada. Unpublished PhD dissertation, Hebrew University of Jerusalem.
Boudana, Sandrine (2015) “Le spectateur engage”: French war correspondents’ conceptions
of detachment and commitment. Journal of European Studies 45(2), 137–151.
Boudana, Sandrine, & Segev, Elad (2017a) The bias of provocation narratives in international
news. International Journal of Press/Politics 13(4), 401–428.
Boudana, Sandrine & Segev, Elad (2017b) Theorizing provocation narratives as communication
strategies. Communication Theory 27(2), 329–346.
Brandes, U., &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In M.
Jünger & P. Mutzel (Eds.), Graph drawing software (pp. 321–340). New York: Springer-Verlag.
Campbell, Joseph (1972) Myths to Live by. New York: Penguin/Arkana.
Catalogue for the Guidance of Foreign Investment Industries. (2004, December 1). Ministry
of Commerce People’s Republic of China. http://english.mofcom.gov.cn/aarticle/poli
cyrelease/gazette/200505/20050500093692.html.
Choi, Jinbong. (2010). The representation of North Korean national image in national
newspapers in the United States. Public Relations Review, 36, 392–394. doi:10.1016/j.
pubrev.2010.08.001.
Chung, M. Y., Lessman, J. R., & Fan, M. (2014). Framing the Nuke: How news media
among countries in the Six-Party talks framed North Korea's nuclear test. North Korean
Review 10(1), 22–38.
Cohen, Bernard C. (1963). The Press and Foreign Policy. Princeton, NJ: Princeton University
Press.
Entman, Robert M. (1993). Framing: Toward clarification of a fractured paradigm. Journal of
Communication 43(4), 51–68.
Entman, Robert M. (2003). Cascading activation: Contesting the White House’s frame after
9/11. Political Communication 20(3), 415–432.
Gans, Herbert J. (1979). Deciding What’s News: A Study of CBS Evening News, NBC Nightly
News, Newsweek, and Time. New York: Pantheon Books.
Graber, Doris A. (1989). Mass Media and American Politics. Washington, DC: CQ Press.
Hackett (1984). Decline of a paradigm? Bias and objectivity in news media studies. Critical
Studies in Mass Communication 1(3), 229–259.
Hall, Stuart, Critcher, Chas, Jefferson, Tony (1978). Policing the Crisis: Mugging, the State, and
Law and Order. London: Macmillan Publishers.
Hallin, Daniel C. (1987). Hegemony: The American news media from Vietnam to El Salvador.
In: David Paletz (ed.), Political Communication Research. Norwood, NJ: Ablex Publishing
Corporation.
Handley, Robert L., & Lou Rutigliano (2012). Journalistic field wars: Defending and
attacking the national narrative in a diversifying journalistic field. Media, Culture & Society
34(6), 744–760.
Herman, Edward S. (1993). The media’s role in U.S. foreign policy. Journal of International
Affairs 47(1), 23–45.
Hinck, Robert, & Cooley, Skye C. (2020). North Korean media penetration and influence
in Chinese and Russian media: Strategic narratives during the 2017‒2018 nuclear con-
frontation. International Journal of Communication 14, 1331–1352.
Iyengar, Shanto (1990). Framing responsibility for political issues. The case of poverty. Poli-
tical Behavior 12(1), 19–40.
70 Y. Zhou, S. Boudana, and E. Segev

Kim, Kyung Hye (2014). Examining US news media discourses about North Korea: A
corpus-based critical discourse analysis. Discourse and Society 25(2), 221–244. doi:10.1177/
0957926513516043.
Knobloch-Westerwick, Silvia, & Taylor, Laramie D. (2008). The blame game: Elements of
causal attribution and its impact on siding with agents in the news. Communication Research
35(6), 723–744.
Lee, Chin-Chuan, & Junghye Yang. (1995). Foreign news and national interest: Comparing
U.S. and Japanese coverage of a Chinese student movement. International Communication
Gazette 56(1), 1–18.
Lu, Yan (2013) Huanqiu Shibao shi zenyang liancheng de [How the Global Times was
tempered]. Fenghuang Zhoukan [Phoenix Weekly], 484, 6–27. https://read.douban.com/rea
der/ebook/4820479/.
Malek, Abbas (1997). Introduction: News media and foreign policy: a field ripe for research.
Journal of International Communication 4(1), 1–10.
McClure, John, Allen, Michael W., & Walkey, Frank (2010). Countering fatalism: Causal
information in news reports affects judgments about earthquake damage. Basic and Applied
Social Psychology 23(2), 109–121.
Meyer, Anneke (2010). “Too drunk to say no” Binge drinking, rape and the Daily Mail.
Feminist Studies 10(1), 19–34.
Meyers, Marian (1994). News of battering. Journal of Communication 44(2), 47–63.
Paletz, David, & Entman, Robert M. (1981) Media, Power, Politics. New York: The Free
Press.
Peters, Chris (2011). Emotion aside or emotional side? Crafting an “experience of involvement”
in the news. Journalism 12(3), 297–316.
Powlick, Philip J., & Katz, Andrew Z. (1998). Defining the American public opinion/foreign
policy nexus. Mershon International Studies Review 42(1), 29–63.
Provisions for the Administration of Internet News Information Services (2017, June 1).
Revision of the provisions for the administration of internet news information services
2005. https://wilmap.stanford.edu/entries/provisions-administration-internet-news-informa
tion-service-2017.
Qin, Wenfeng, & Wu, Yanyi (2019). jiebaR: Chinese text segmentation. R package version
0.11.
Robinson, Piers (1999). The CNN effect: Can the news media drive foreign policy? Review
of International Studies 25(2), 301–309.
Robinson, Piers (2002). The CNN Effect: The Myth of News Media, Foreign Policy and Inter-
vention. New York: Routledge.
Segev, E. (2016). International News Flow Online: Global Views with Local Perspectives. New
York: Peter Lang.
Seo, Soomin, & Nah, Seungahn (2020). Mapping communication research concerning
North Korea: A systematic review (2000–2019). International Journal of Communication 14,
1308–1330.
Sigal, Leon V. (1973). Reporters and Officials: The Organization and Politics of Newsmaking.
Lexington, KY: DC Heath and Company.
Tilleman, Morgan (2010). (Trans)forming the provocation defense. Journal of Criminal Law
and Criminology 100(4), 1659–1688.
Wall, Irvin (2008). France in the Cold War. Journal of European Studies 38(2), 121–139.
Welch, Susan (1972) The American press and Indochina. 1950–1956. In: Richard L. Merritt
(Ed.), Communication in International Politics. Urbana. IL: University of Illinois Press.
Wu, Guoguang (1994). Command communication: The politics of editorial formulation in
the People’s Daily. The China Quarterly 137, 194–211.
 Provocation narratives 71

Yanxi (2017, December 5). Duizhun chaoxian 700 mubiao meiti caice zhunbei junshi
fangan chaoxian cheng meihan junyan shi hezhan qianzou [Military exercises aim at 700
North Korea targets, media speculates about military action plan, North Korea proclaims
the prelude of nuclear war].Huanqiu Shibao [Global Times], 16.
Zaller, John R., & Chiu, Dennis (2000). Govemment’s little helper: U.S. press coverage of
foreign policy crises, 1946–1999. In B. Nacos, R. Shapiro, & P. Isemia (Eds.), Decisionmaking
in a Glass House (pp. 61–84). Lanham, MD: Rowman & Littlefield.
Zhan, Debin (2016). Analysis of changes in North Koreas cognition of China through its
media coverage. Korean Journal of Defense Analysis 28(2), 199–221.
Zhang, Weiqi, & Zinoviev, Dmitry (2018). How North Korea views China: Quantitative
analysis of Korean central news agency reports. Korean Journal of Defense Analysis 30(3),
377–396.
4
CABLE NEWS CHANNELS’ PARTISAN
IDEOLOGY AND MARKET SHARE
GROWTH AS PREDICTORS OF SOCIAL
DISTANCING SENTIMENT DURING
THE COVID-19 PANDEMIC
James A. Danowski, Bei Yan and Ken Riopelle

The return of media partisanship

In times of heightened polarization, it is not surprising that the US public’s adoption
of methods to mitigate the spread of the COVID-19 virus, such as social distancing
and related measures, became politicized. Liberals became more likely to practice
mitigation, while conservatives less so (Rothberger et al., 2020; Hart et al., 2020;
Christensen et al., 2020). This pattern was also observed in other countries, such as
New Zealand (Becher, et al., 2020), Brazil (Ramos et al., 2020), and the UK
(Harper & Rhodes, 2020).
Because the commercial media strive to increase audience share to raise advertising
fees, they tend to produce news content to attract partisan audiences. In 2020,
observers viewed the media as rising in partisanship in the climate of heightened
polarization (Jurkowitz et al., 2020). Yet before considering the contemporary media
partisanship climate, its historical trajectory merits a brief treatment.
The term “partisan press” refers to journalism that is systematically influenced
and sometimes affiliated with the government and its parties. The frequency of the
bigram “partisan press” is portrayed in Figure 4.1 using Google Books1 of Amer-
ican English published since the 1800s. It indicates that this concept has gained a
renewed interest in the 21st century. The period of highest turbulence, from 1820
to 1890, declined with the Progressive Era of the 1890s–1920s, further declined
during World War II and the post-war period, and then increased in the 1980s to
its highest levels since the turn of the previous century.
Data source: Google Books Ngram Viewer (https://books.google.com/ngrams)
As the 1800s unfolded in the post-Revolutionary War period, political parties
funded newspapers, as hundreds grew to a thousand and more by the 1830s
(Formisano, 2008). However, while press partisanship increased, most readers
were unaware of this undercurrent shaping the stories on the surface (Baughman,

DOI: 10.4324/9781003120100-5
 Partisan ideology and market share growth 73

FIGURE 4.1 “Partisan Press” mentions in Google books since the 1800s.

2011). The abolitionist movement further fueled partisanship, leading up to the

Civil War. The Progressive Era’s investigative journalism of the 1890s–1920s saw
a critical, independent press emerge alongside the Democratic- and Republican-
funded newspapers. When muckraking investigative journalism emerged, so did
the reporting on corporate, union, and governmental corruption, and the result-
ing social injustices. The attention to these areas appears to have dampened the
salience of press partisanship in that period.
Nevertheless, by the 1940s, issues emerged about some radio stations’ biased
coverage of local politics. A Federal Communications Commission policy formed
that radio stations served the public interest and therefore could not take editorial
positions on issues, a policy further articulated in the Fairness Doctrine in 1949
(Simmons, 1978) that required broadcasters to cover controversial issues and pre-
sent opposing viewpoints, a sort of common carrier opinion model. At the same
time, the US Marshall Plan for European recovery (Hogan & Hogan, 1987)
reflected this policy but with a different implementation. To reduce the likelihood
that a one-sided press with no opposition would re-emerge, the program funded
diverse political parties to establish viable media outlets, resulting in a more
balanced overall media domain.
Domestically, the Fairness Doctrine guided media coverage during the 1950s and
early 1960s. The social movements of the 1960s–1970s appeared to increase media
partisanship, perhaps because counterculture supporters’ liberal orientations fit with
journalistic norms. This may have contributed to the erosion of the Fairness Doctrine,
as it ceased to be enforced and officially abandoned in 1987.
Another factor increasing media partisanship was the emergence and growth of
cable news outlets, with CNN launched in 1980. It featured a popular primetime
show, Crossfire, which pitted a conservative against a liberal commentator, an exten-
sion of the two-sided approach to media coverage fostered by the soon-to-be-defunct
Fairness Doctrine. Then as Fox News and MSNBC emerged 16 years later, they
tended toward a more partisan treatment, favoring one side over the other. CNN was
ideologically positioned between them but closer to the liberal partisan perspective. In
74 J. A. Danowski, B. Yan, and K. Riopelle

2020, quantification of partisanship in the Gallup/Knight study, American Views

2020: Trust, Media, and Democracy, scaled MSNBC as left with a score of 1, CNN as
left-center at 1.25, and Fox News as conservative at 4.75 on a five-point scale.
Although media partisanship is widely considered as a common practice in the US
of 2020s, coloring coverage, the motive to attract larger audiences may interact with
the sentiment expressed in the news toward issues of the day. When sentiment
increases, both positive and negative, it heightens audience engagement (Arapakis et
al., 2014), which leads to increased viewership. Accordingly, the market motive may
be a stronger factor than ideology in channeling sentiments. Perhaps the rising
interest of scholars in media partisanship results from the convergence between
ideology and market logic. With the growing media fragmentation, the sensational
partisan press sells more.
Nevertheless, media partisanship growth raises questions about whether ideological
perspectives have taken a deeper root in infusing news content production or is a
more transitory market-based orientation, capitalizing on differences in audience
member’s political orientation. Has the media profit-seeking model, which symbio-
tically links their media content with audiences’ political perspectives, been altered by
a new threshold of ideological framing? Have the media become a more active agent
of change than the common carrier of news before the Fairness Doctrine’s dissolu-
tion? Has the revenue-driven media become more the cause of partisan polarization
than a reflector of it? Has ideology today become yet another commercial brand?
Critical media theory (Schiller, 1991, 2013; Fuchs, 2011) would agree with this
view, stressing that capitalist ideology frames and shapes media content. These two
perspectives, one that partisan ideology accounts for social distancing sentiment, and
the other that market share growth is stronger, is the basis for the two hypotheses
addressed in this chapter. One is that ideology is the primary determinant of media
sentiment toward social distancing, and the other is that market motives offer a better
explanation. Over time, partisan ideology may have lost some of its meaning as it
dissolved into the market logic. This would explain why we have seen a sharp
decline in partisan media during the 20th century and now again a rise in attention
in the partisan press. It may be that partisan press and market forces today do not
compete as they did in the 20th century but rather complete each other.

The rise of social distancing and its political context

Merriam-Webster states that social distancing is a medical term, first used in 2003.
However, as early as 1972, the scholarly literature used the term to refer to a dif-
ferent sense, the physical and social distance between social units, in this case tea-
chers and students (Schwebel & Cherlin, 1972). Yet the concept of social
distancing has a much longer history. Removing sick individuals from social set-
tings to prevent the spread of disease dates back at least to 538 BC, when the
Bible’s Book of Leviticus was first written, describing removing lepers from the
camp. Issac Newton isolated himself from the disease, and lepers were exiled to
colonies on Molokai to prevent the disease spread to other Hawaiians.
 Partisan ideology and market share growth 75

Imagine the leper’s distress in being forcefully excluded from the social net-
work with no hope of return, a separation more extreme than experienced
during the recent governmental campaigns, premised on a much shorter time
horizon until returning to normal. Nevertheless, the adverse effects attributed to
social distancing during the pandemic are significant: increased mental health
problems, substance abuse (Panchal et al., 2020), suicide (Thakur & Jain, 2020),
and domestic violence (Campbell, 2020). Moreover, participation in violent
protests during the pandemic may have increased due to social distancing. These
negative effects linked to social distancing offer unique evidence of the social
network’s importance in maintaining normative behavior. When the network is
suppressed, negative consequences soon arise.
The social network’s central role makes it difficult to promote social distan-
cing, even on a short-term basis, so information campaigns require highly per-
suasive messages with many repetitions. The initial warrants in the argument
were that social distancing would protect vulnerable elderly, a more collective
intergenerational appeal. Not until later in the campaign did the rationale shift
toward more individualistic protection of the self, as masks were advocated for
the general public in addition to the medical providers, not only as a means of
protecting others. Once the arguments reached this level, perhaps the altruistic
motive dissipated, as individuals weighed their risks against the somatic limita-
tions of mask-wearing and loss of personal freedom and practiced social dis-
tancing less.
These social distancing information campaign messages are filtered through par-
tisan media biases. When the coronavirus campaign launched, daily press briefings
from the White House were covered by the three cable channels, but by April 1,
2020, CNN and MSNBC ceased broadcasting the briefings (Wemple, 2020), while
Fox News continued coverage until the daily briefings ended on April 26. More-
over, as the media reported on compliance, this likely reinforced the respective
attitudes and behaviors regarding social distancing.
The embracing of social distancing by the liberal press but not the conservative
press may be due to the media channels attention to the base reaction of mem-
bers of the public based on their partisan preferences, where conservatives more
likely see social distancing as limiting personal freedom, while liberals would see
more of the positive collective benefits of social distancing. A more contemporary
current influencing media sentiment may be the fact that President Donald
Trump was portrayed as somewhat negative toward social distancing as the
COVID-19 campaign progressed and more concerned with the economic
implications of the pandemic, while presidential candidate Joe Biden was more
positive toward social distancing. In this immediate partisan election climate, the
cable channels were aligned in apparent support of one candidate over the other.
The fact that the same liberal/conservative stances toward social distancing were
reported in other countries would reduce the likelihood that the social distancing
sentiment was primarily based on election sentiment that media expressed during
the period.
76 J. A. Danowski, B. Yan, and K. Riopelle

The partisan profile of cable news channels suggests the hypothesis that Fox
News would be more negative toward social distancing, while MSNBC and CNN
more positive. An alternative hypothesis is that market share growth predicts social
distancing sentiment better than ideology, which suggests that CNN, with the
largest increase in market share, should have the most sentiment expressed, fol-
lowed by Fox News, and MSNBC.

Methods

Sentiment measurement model2

The dominant methods for sentiment analysis (Kharde & Sonawane, 2016) seek to
classify messages as positive or negative for use in machine or deep learning using
neural network models (Zhang et al., 2018). Less common are methods that mea-
sure the degree of positivity or negativity in texts. Classification of textual content
into positive or negative categories (Liu & Zhang, 2012; Mäntylä et al., 2018)
counts frequencies of sentiment words in a lexicon, a predefined list, or dictionary
of positive and negative terms. Counting individual word frequencies is referred to
as a “bag-of-words” model. The approach treats all of the words in textual units of
observation disaggregated and jumbled together with no relations among them.
The proximity of words in the text is ignored. The bag-of-words sentiment scores
are typically based on counts of binary values for whether each word in the text
appears in a document.
In communication science, rather than classification, content analysis of mes-
sages to measure the degree of positive and negative sentiment associated with a
target is often the goal. This content analysis requires a different measurement
model than bag-of-words, one based on a network approach. Although most
social network analyses are of relationships among entities, such as individuals,
groups, organizations, or nations (Rogers, 1987; Monge and Contractor 2003;
Borgatti et al., 2009), a network model has also been useful in treating words in
the text as nodes and their proximate co-occurrences as links, forming a semantic
network (Danowski, 1982, 1993; Carley, 1993; Corman et al., 2002).
Some recent examples of semantic network analysis include work by
Danowski and Park (2014), Jiang et al. (2016), and Danowski and Riopelle
(2019). Semantic network analysis covers a wide range of meaning aspects
(Osgood et al., 1957). An essential advantage of semantic network analysis is that
it illustrates the relationships among words, thus generating insights about the
entire text’s structures and meanings. Here we are concerned not only with the
sentiment, which is just one dimension among many that semantic network
analysis can index in the study of texts. Nevertheless, we present a sentiment
analysis approach building on word relationships and embeddedness in texts. This
method can potentially be applied to other dimensions of texts, as long as
researchers are interested in looking for the strength of relationships between a
target word or phrase and a particular category of words.
 Partisan ideology and market share growth 77

Approaches to sentiment analysis

Lexicon-based measures
The most simple and common approach for sentiment analysis is using a predefined
lexicon or dictionary containing sentiment words, affective orientations, and some-
times the strength of its orientation. Following the bag-of-words approach, lexicon-
based approaches first break down a body of text into independent words. Then it
counts the frequency of sentiment words (which are defined by the lexicon used)
that appear in the text and computes a sentiment score of the text, usually in the
form of a percentage based on the word count.
Commonly used sentiment lexicons include Linguistic Inquiry and Word Count
(LIWC) (Tausczik & Pennebaker, 2010), SentiWordNet (Baccianella et al., 2010),
and the Bing lexicon. Whereas some lexicons, such as Bing, contain words in
binary categories, others, like SentiWordNet, provide a ratio indicating the words’
orientation’s strength. Lexicon-based sentiment analysis is an unsupervised method
that is easy to apply and not domain-dependent. It can be highly accurate if used
appropriately (Kundi et al., 2014; Asghar et al., 2014; Khoo & Johnkhan, 2018).
However, a limitation of lexicon-based measures based on the bag-of-word
approach is that it focuses only on the frequency of single, tokenized words. It
omits the words’ contexts based on their co-occurrences in the texts that are critical
to sense-making. In other words, you get one score for an entire text regardless of
the number of persons, organizations, or brands mentioned.

Machine learning classification

The machine learning approach utilizes supervised learning, which starts by
extracting features from texts (Liu, 2012). Machine learning algorithms applying
the bag-of-words approach treat single words as semantic features. The features
and outcomes (e.g., annotated sentiment of texts) it learns from the training text
data classifies texts into different sentiment categories, such as positive, negative,
and neutral. The key to the performance of machine learning lies in the effec-
tiveness of the features it extracts. The machine learning approach has the edge
over the lexicon-based measures as it acquires information directly from the text
body rather than a standard lexicon (D’Andrea et al., 2019. Therefore, it is better
customized to the text data.
Several software vendors, including IBM (Watson), Google (Cloud Natural
Language), Amazon (Comprehend), and Microsoft (Azure), have developed their
own proprietary machine learning algorithms for sentiment analysis. These algo-
rithms are relatively easy to use but are not transparent (since they are proprietary)
and can be expensive for researchers. This is because a machine learning-based
sentiment analysis can be costly to develop. It requires a considerable amount of
text data to train an accurate classification algorithm and may need human coders
to annotate the training texts. It may also work better for long documents than
78 J. A. Danowski, B. Yan, and K. Riopelle

short reviews or tweets so that there are more words to serve as textual features for
classification (Khoo & Johnkhan, 2018). Machine learning classification using a
bag-of-words approach also shares the same limitation with the lexicon-based
approach. It only uses independent words and ignores the contexts in which the
words are embedded.

Word embeddings for sentiment analysis

A relatively recent development in natural language processing is word
embeddings (Mikolov et al., 2013). Word embeddings are techniques that map
words in a text into numeric vectors in a vector space. Instead of assuming
words as independent, as the bag-of-words approach does, word embeddings
often operate based on a sliding window and extract features from a sequence
of words cooccurring in a body of text. This approach aligns with the semantic
network perspective and takes into consideration word contexts. Based on how
words appear with one another, word embedding algorithms represent the
words in the vector space, in which words used in similar ways are closer to
one another.
Word embeddings may be applied in two ways in sentiment analysis. The first
is by extracting words and their relations in the texts as features for sentiment
classification (Kumar & Zymbler, 2019). Researchers have also applied pre-
trained word embedding corpora to classify texts. So, when target texts contain
words that did not appear in the training dataset, the algorithm can judge text
sentiment based on how close the new words are to the words that appear before
in the relational corpora (Rudkowsky et al., 2018). Just like other machine
learning models, training with word embeddings requires the dataset to be large
to produce an accurate mapping of words in a text. If pre-trained word embed-
ding corpora are used, then the algorithm does not directly learn from the body
of text being analyzed and may not precisely capture the texts’ local context
under scrutiny.

Aspect-based sentiment analysis

Based on the unit of analysis, sentiment analysis can also be classified as either
subjectivity/objectivity identification or feature/aspect based. Subjectivity/
objective identification, as used by studies cited above (e.g., Kumar & Zymbler,
2019; Rudkowsky et al., 2018), classifies the sentiment of an entire text. By
contrast, aspect-based sentiment analysis takes a more fine-grained approach,
aiming to determine sentiment in parts of texts (e.g., opinions regarding dif-
ferent attributes of a product or service) (Pontiki et al., 2016; Thet et al., 2010;
Wang & Liu, 2015). For example, when analyzing an online review of a hotel,
the subject/objective identification estimates the review’s general sentiment. In
contrast, the aspect-based sentiment analysis may examine how positive the
review is toward the hotel’s location, service, room, and food. Therefore, the
 Partisan ideology and market share growth 79

first step of aspect-based sentiment analysis involves parsing texts into different
linguistic components through automated algorithms such as topic modeling
(Thet et al., 2010). After the texts are broken down, researchers can then
choose to apply the sentiment analysis discussed above to measure the aspect-
specific sentiments. Aspect-based sentiment analysis thus provides more detailed
and accurate information regarding the sentiment in texts, which can be parti-
cularly useful when one needs to understand opinions about specific features.
In summary, existing sentiment analysis methods commonly apply the bag-of-
word approach, breaking texts down to independent words without considering
word contexts. The more recently proposed word embeddings approach is
gaining traction, but machine learning using the method requires a large
amount of data. Using pre-trained word embeddings makes judgments based on
previously collected data rather than the texts being analyzed. Therefore, it risks
missing critical information in the local word context. The semantic network-
based approach to sentiment analysis proposed in the current study comple-
ments the above methods. It overcomes the limitations of the bag-of-words
model by gauging the contexts of words in texts based on word sequence and
co-occurrence. It has an advantage over machine learning approaches as it does
not need a large amount of data and measures sentiments based on the local
information in a given text.
Moreover, the semantic network approach allows fine-grained sentiment analysis
at the aspect or feature level like aspect-based sentiment analysis. Instead of relying
on unsupervised learning algorithms such as topic modeling to identify features in a
text, this approach enables researchers to name the target word or phrase of interest
(person, organization, event, or brand, e.g., iPhone). It generates a score indicating
sentiment toward this specific target. Thus, the sentiment network method can
generate sentiment scores for multiple targets of interest in the same text, enabling
a comparison of the results.
The sentiment network approach measures target-specific sentiment based on
the shortest paths between the semantic network’s target and sentiment words.
The method has three significant advantages over bag-of-words classification
approaches: 1) the network method measures sentiment concerning targets,
which is possible because the basic unit of analysis is the word pair in a sentence,
not a document; 2) the more micro-level word pairs are links in a chain, forming
shortest paths that extend across text units, enabling tracing the closeness of sen-
timent words to a target word or phrase; and 3) the sentiment network approach
can compare multiple targets in the same corpus, which expands the scope of
testable hypotheses.
The software we used to measure sentiment in this fashion, SENET, is described
in greater detail in Appendix B (see eResources), including the code in R. It covers
greater detail on the computational aspects of the current study for those who may
wish to explore the tools further: 1) data acquisition from GDELT, 2) semantic
network analysis using WORDij, 3) group detection and graphing in NodeXL,3
and 4) sentiment network analysis with SENET.
80 J. A. Danowski, B. Yan, and K. Riopelle

Data
Data for this study included 150-character snippets of news content containing the
term social distancing from television transcripts from Fox News, MSNBC, and
CNN from January through September 2020, produced by GDELT (https://
gdeltproject.org). Here are examples of snippets:

8/20/2020 13:22 CNN … were on the strictest lockdown in the entire

nation for
approximately i believe 15 weeks. many people complied with wearing masks,
practicing physical distancing. i know that we were social distancing, but we
were practicing physical distancing and people took care of one …
8/20/2020 18:22 CNN … touches on their convention. thanks very much.
appreciate it.
los angeles shutting off power to a mansion that’s been holding parties, despite
social distancing rules, and they monitored to …
8/20/2020 5:14 CNN … he said i’ve been quarantining, self-isolating, social
distancing
for the past six months, but this election is important enough, i believe it’s
important enough for me, to go to the …

After pre-processing text to remove numbers, punctuation, and stop-words, we

conducted semantic network analysis with WORDij software (http://WORDij.
net). No stemming was performed to enable capturing sentiment nuances and a
higher fidelity representation of meaning.
Media partisanship scores were obtained from a 2020 Gallup/Knight study on
media trust and democracy.4 Media partisanship was scaled based on the coding
schemes of Media bias/Fact check5 and Allsides.6 MSNBC was rated as left at 1.0,
CNN as left-center at 1.25, and Fox News as conservative at 4.75.
Market share growth refers to an increase in audience share during the pandemic in
2020. The data were reported (Schneider, 2020) in Variety.7 Table 4.1 shows the
market share growth of the three cable outlets. CNN has the largest growth, followed
by Fox News and CNN. Market share is in millions of viewers as of January 1, 2020.
We measured positive and negative sentiment toward “social distancing” with
the SENET semantic network sentiment analysis package in R (Danowski et al.,
2020). The procedure begins with the creation of the word co-occurrence net-
work based on the sliding window that tracks the appearance of pairs of words in

TABLE 4.1 Market share and growth for Fox News, MSNBC, and CNN

Cable News Channel Share Growth

Fox News Channel 2.501 +43%
MSNBC 1.741 +23%
CNN .965 +83%
 Partisan ideology and market share growth 81

it, cumulating the counts. With this network, we then use a lexicon of positive and
negative words developed through the compilation of the lexicons of others and
measure the distance to and from each sentiment word to the target word or phrase, in
this case “social distancing”, by tracing the shortest paths linking them. Then we
inverted these values, squared them, and multiplied them by the co-occurrence fre-
quencies along the path. Appendix B (see eResources) describes this in more detail,
including the R code to compute the sentiment scores.

Results
Figure 4.2 shows the volume of coverage of social distancing. Data for January and
February 2020 do not appear because there were fewer than five references. On
March 16, the federal government introduced the first campaign, 15 Days to Slow
the Spread. The mentions of social distancing increased to a peak in April 2020,
followed by a decline toward June 2020. Figure 4.2 further shows that the same
pattern occurs for negative and positive sentiment. CNN has the most negative and
positive sentiment expressed. Comparing the curves for the frequency of coverage
and sentiment suggests an association between the frequency of coverage and sen-
timent, which are consistent with findings of earlier research (Danowski & Rio-
pelle, 2019) that sentiment produces an increase in the volume of coverage. As
well, there appears to be a relationship between positive and negative sentiment. As
one increases so does the other, although the differences between them over time
show considerable variability in this association, particularly as seen for April and
August. Although the coverage of Fox News, MSNBC, and CNN are parallel for

FIGURE 4.2 Social distancing frequencies and sentiment for Fox News, MSNBC, and
CNN.
82 J. A. Danowski, B. Yan, and K. Riopelle

most of the duration. In August there is a deviation where CNN has higher sen-
timent than Fox News and MSNBC, both of which decline, while CNN increases
its expression, particularly of positive sentiment.
The key analysis tests the ideology vs. the market share hypothesis for the overall
expression of sentiment (both positive and negative). Figure 4.3 has the y-axis
normalized for the two variables. It shows evidence for a linear relationship
between market share growth and the amount of sentiment expressed (right), while
for ideology and total market share (left), this is not the case. The lines cross. This
supports the market share growth hypothesis over that of ideology.
The finding that August had the greatest deviation among the cable news channels
motivated our analysis of the word co-occurrences for the outlets. Because MSNBC
did not have coverage above the lower frequency threshold of three co-occurrences,
its network is null, and the comparison is between CNN and Fox News. The bigram
“social distancing” was converted to the unigram “socialdistancing” to create a
semantic target to represent the concept; otherwise, targeting social or distancing
would introduce much measurement error. We examined the data to see the differ-
ences in word pair co-occurrences, shown in Tables 4.2 and 4.3. The differences
found motivated an analysis of the overall semantic networks for the two channels.
CNN’s co-occurrence frequencies were 5.9 times higher than Fox News’s,
which produced a difficult to comprehend hairball network graphic, so to sparsify

FIGURE 4.3 Ideology, market share, and market share growth vs. social distancing
sentiment.
 Partisan ideology and market share growth 83

TABLE 4.2 Higher August word pair frequencies for CNN

WORD PAIR CNN FOX % % Z-

FRQ FRQ SCORE
masks socialdistancing 312 0 0.0195 0.0000 6.92
no socialdistancing 194 0 0.0121 0.0000 5.44
not socialdistancing 180 0 0.0112 0.0000 5.24
socialdistancing masks 156 0 0.0097 0.0000 4.88
wearing socialdistancing 142 0 0.0089 0.0000 4.65
mask socialdistancing 118 0 0.0074 0.0000 4.24
socialdistancing wearing 113 0 0.0071 0.0000 4.14
no masks 111 0 0.0069 0.0000 4.11
socialdistancing not 106 0 0.0066 0.0000 4.01
people socialdistancing 95 0 0.0059 0.0000 3.80
white house 78 0 0.0049 0.0000 3.44
socialdistancing mask 67 0 0.0042 0.0000 3.19
people wearing 101 3 0.0063 0.0012 3.10
right now 62 0 0.0039 0.0000 3.07
masks social 55 0 0.0034 0.0000 2.89
wearing masks 280 23 0.0175 0.0095 2.88
ensure socialdistancing 54 0 0.0034 0.0000 2.86
masks required 53 0 0.0033 0.0000 2.83
masks distancing 52 0 0.0032 0.0000 2.81

TABLE 4.3 Higher August word pair frequencies for Fox News

WORD PAIR CNN FOX % % Z-

FRQ FRQ SCORE
health people 11 6 0.0000 0.0281 –21.26
should masks 11 6 0.0000 0.0277 –21.10
people together 14 7 0.0000 0.0273 –20.94
lot people 38 14 0.0000 0.0273 –20.94
think people 13 9 0.0000 0.0273 –20.94
students socialdistancing 0 10 0.0000 0.0215 –18.58
without socialdistancing 0 10 0.0000 0.0211 –18.40
wear socialdistancing 0 11 0.0000 0.0202 –18.03
no socialdistancing 0 12 0.0000 0.0202 –18.03
socialdistancing guidelines 0 12 0.0000 0.0202 –18.03
socialdistancing mask 0 13 0.0000 0.0153 –15.67
socialdistancing wearing 0 14 0.0000 0.0136 –14.79
practicing socialdistancing 0 15 0.0032 0.0289 –14.62
socialdistancing people 0 15 0.0000 0.0124 –14.10

(Continued)
84 J. A. Danowski, B. Yan, and K. Riopelle

TABLE 4.3 Cont.

WORD PAIR CNN FOX % % Z-
FRQ FRQ SCORE
mask socialdistancing 0 18 0.0000 0.0124 –14.10
not without 0 18 0.0000 0.0112 –13.38
wearing socialdistancing 0 27 0.0000 0.0074 –10.92
socialdistancing masks 0 30 0.0000 0.0074 –10.92
socialdistancing not 0 30 0.0000 0.0062 –9.97
face coverings 51 70 0.0000 0.0062 –9.97

Note. The headings in this table are as follows: FRQ is the word co-occurrence frequency, % is the
proportion based on the total number of words for CNN and Fox News, respectively. Only values of p
< .001 are displayed.

the network we set the lower frequency threshold at 10. The size of nodes in the
graphs is based on betweenness centrality (Brandes, 2001). Figure 4.4 shows the
networks of CNN and Fox News. The main substance of the two is seen in Tables
4.4 and 4.5, which show the top ten words in the top seven groups in each. Word
groups were identified using the Clauset-Newman-Moore community detection

FIGURE 4.4 Social distancing semantic network CNN August 2020.

 Partisan ideology and market share growth 85

FIGURE 4.4 (cont.)

algorithm (Clauset et al., 2004). This method identifies communities by finding the
node cluster that produces the largest increase in modularity, or the relative density
of edges within communities to those outside communities. There are other faster
clustering algorithms, such as the Louvain method (Blondel et al., 2008). The
reason we chose the Clauset-Newman-Moore algorithm is that it is available in
NodeXL, which we used to generate the networks.
The word groups in CNN’s August 2020 coverage, containing at least three
words, numbered 24. Table 4.4 lists the seven largest groups, including the top ten
most central words in each. Table 4.5 shows the word groups for Fox News. It had
only seven groups, and they occur at considerably lower frequencies. The order of
groups reflects their size. For comparison purposes, the top seven largest groups are
shown for CNN. Because our interest is not in mapping all the meanings for social
distancing on CNN, but only to contrast the main differences between CNN and
Fox News, we list the top seven largest groups for CNN, rather than all 24.
CNN’s largest word group is about people not wearing masks or social distancing,
while Fox News reinforces guidelines for it. So, while CNN is pointing out violations,
Fox News is promoting social distancing. In downplaying lack of compliance, Fox
News exhibits conservative ideology, yet at the same time encourages social
86 J. A. Danowski, B. Yan, and K. Riopelle

TABLE 4.4 CNN August word groups

1 2 3 4 5 6 7
possible avoid people Reopening really health day
masks large risk States president dont visit
not gatherings putting Seeing event defying pay
enforced crowds taking new heads state daughter
no socialdistancing sick coronavirus mt officials kids
wearing mandatory trying cases trumps experts back
think steps attend today trump public school
wash practices thousands ways supporters system youre
mask making expected jersey rally care time
wear way gathered yorkers donald safety put

TABLE 4.5 Fox News August word groups

1 2 3 4 5 6 7
socialdistancing people practicing health good safe lot
follow close wearing guidelines extra engaging now
operation together not no sanitizing really inside
allowed coronavirus mask public checks common tech
protocols masks testing ignoring face measures campaigns
concern things wear pandemic coverings
cases think house lives
miss hand sound own
numbers required doesnt way
outbreak covid situation good

distancing, a stance considered more liberal. This is further evidence for the failure of
ideology to explain the coverage of social distancing. More consistent with the data is
marketing motives. By not expressing as much negative sentiment toward lack of
compliance, they avoided risking their audience loyalty for the bulk of its audience,
conservatives. Yet, by encouraging social distancing, they also appeal to more liberal
audience members. Fox did use the slogan “Fair and Balanced”, and during daytime
hours included opposing partisan commentators in panels. Thus, consistent with the
findings from the comparison of ideology vs. market share growth as predictors of
sentiment, here among the whole semantic networks, we found further evidence at
the level of the substantive content showing the failure of ideology to explain media
coverage of social distancing.
In CNN's August coverage, the first word group is about not wearing masks. The
second is about the need to avoid large gatherings. Group 3 is about the Southern
 Partisan ideology and market share growth 87

California beach crowds, while Group 4 is about reopening and a surge in cases.
Group 5 is about Trump rallies, while Group 6 is about defying state public health
officials, and Group 7 is about schools.
The top word groups in August for Fox News appear in Table 4.5. Group 1 is
about social distancing protocols and numbers of cases, while Groups 2 and 3 are about
the need to wear masks, Group 4 is about the public ignoring health guidelines, Group
5 is about sanitizing and face coverings, Group 6 is about engaging in safe common
measures, while Group 7 is about tech campaigns.
Table 4.6 shows the top 20 negative and positive sentiment words that appeared in
CNN and Fox News’s coverage in August. The values range between 0 to 8, based on
the frequency weighted inverse square of the shortest paths linking with social distan-
cing, representing the closeness and strength of the target to the sentiment words.

TABLE 4.6 CNN August social distancing sentiment word strengths

CNN Fox
NEGATIVE POSITIVE NEGATIVE POSITIVE
problem 8.00 celebrate 8.00 no 7.00 good 6.00
difficult 7.00 effective 8.00 not 6.00 engaging 3.00
lose 6.00 working 7.00 concern 4.00 new 3.00
hard 5.00 ready 6.00 disgusting 4.00 safe 3.00
crowded 4.00 yeah 6.00 miss 3.25 thank 3.00
infections 4.00 absolutely 5.00 ignoring 3.00 young 3.00
issue 4.00 big 5.00 outbreak 3.00 open 2.00
nonexistent 4.00 enjoy 5.00 protests 3.00 significant 1.75
nothing 4.00 great 5.00 risk 3.00 recommenda- 1.50
tions
vice 4.00 proper 5.00 cry 2.56 safely 1.50
controversy 3.00 works 5.00 suspect 1.50 rose 1.00
ill 3.00 young 5.00 worried 1.44 safe 5.00
impossible 3.00 care 4.00 no 8.00 totally 3.25
infection 3.00 clear 4.00 refused 4.00 open 2.50
partisan 3.00 early 4.00 cry 3.75 new 2.25
risk 3.00 encour- 4.00 not 1.75 recommenda- 1.50
aged tions
seriously 3.00 hope 4.00 suspect 1.50 engaging 1.22
terrible 3.00 open 4.00 worried 1.22 significant 1.00
worst 3.00 significant 4.00 ignoring 1.00 young 1.00
dangerous 2.75 thank 4.00 disgusting 0.75 good 0.72
88 J. A. Danowski, B. Yan, and K. Riopelle

In summary, the results for August, in which the sentiment differences between
CNN and Fox News became greatest, were further explored with four methods,
comparing: 1) word pair’s relative frequencies, 2) the overall semantic networks for
the two news channels, 3) cluster analysis, and 4) sentiment analysis. The analysis
comparing significant differences in the two outlets’ word pair frequencies found
CNN emphasizing the lack of social distancing and Fox News attending more to the
health implications. CNN’s overall semantic network had about six times more co-
occurrence frequencies, producing a more developed network, with CNN having
24 groups, while Fox News had only seven. CNN focused more on the lack of
enforcement, while Fox News encouraged social distancing. In terms of negative
sentiment, CNN focused on the problems to enforce social distancing. On the
positive side, it was more exuberant in celebrating the benefits of social distancing.

Discussion
This study used semantic network analysis and sentiment analysis to compare the
discourse on social distancing in three television channels: CNN, Fox News, and
MSNBC. We offered two hypotheses to explain those differences: market share
growth and political ideology. Our findings indicate that market share growth is a
better predictor of social distancing sentiment than ideology. This evidence sup-
ports the theory that sentiment engages audiences and accounts for increased
viewership during the pandemic period studied. The market share dynamics are
consistent with the critical theory. The findings suggest that media partisanship may
be only a strategy for market segmentation and that ideology today is not more
than a brand.
The network-based sentiment model performed as expected in quantifying
positivity and negativity, enabling measurement of sentiment toward a particular
target, rather than the cruder bag-of-words model that measures sentiment at the
whole document level only. Previously shown to have internal and external
validity, the finer-grained network approach yielded more meaningful variation
and evidence, enabling testing the ideology vs. the market share growth hypothesis.
In addition to positivity and negativity scores, the procedures also identified the
positive and negative sentiment words that occurred and the shortest weighted
paths linking them. We mapped the full semantic networks underlying sentiment
and identified the significantly different elements.
The analysis shows how semantic network analysis can be used to test hypoth-
eses even when the number of cases is small, in this case, only three news channels.
Often case studies do not take advantage of the quantification and hypothesis-
testing that semantic network analysis affords. Here semantic and sentiment analyses
provide substantial evidence that blends two traditional approaches, qualitative and
quantitative. The multi-level and multi-method semantic analysis enables the gen-
eration of more knowledge per unit of research effort. The energy no longer spent
on manual coding can be directed to theory development.
 Partisan ideology and market share growth 89

Limitations included a constraint on the availability of data. While, as of this

writing, the data continue to be made available. Due to a technical glitch, the
social distancing snippets ceased as of October 19, 2020, so the current analysis
ends with September’s last full month. Another limitation was on the different time
scales of the data. While snippets are available daily, the market share growth data
and ideological coding for the three channels had single values for the duration, so
the hypothesis test was not based on panel data.
Additionally, our choice of a monthly interval for aggregating the snippets was
most appropriate for examining trends yet had sufficient amounts of text in each
interval for robust semantic network and sentiment analysis. The choice of the
three general 24-hour cable news channels – Fox News with the largest market
share, followed by CNN, and MSNBC, was to control for the news format. We
set aside the outlets with smaller audiences such as BBC, Al Jazeera, Deutsche
Welle, and R.T., the more specialized channels such as Bloomberg, CNBC, and
Fox Business Channel, CSPAN, and the brief news programming of the ABC,
NBC, and CBS networks (PBS transcripts are not available). Future research may
examine these news outlets.

Tips and lessons from the process

1. Analyzing semantic networks of television news requires access to tran-

scripts of broadcasts. Typically, researchers use LexisNexis to obtain the
texts. However, during the pandemic, GDELT provided 150-character
snippets centered on keywords, such as COVID-19, social distancing,
masks, testing, and several others.8 In our research, working with the
GDELT data has required extensive file management to reorganize them
for semantic network analysis. Once these data are no longer available, the
most likely source of television news transcripts is LexisNexis.9 While these
data are useful, the process of downloading data is cumbersome, allowing
only 100 documents at a time. Nevertheless, with persistence, one can
build an adequate corpus of television news texts for US outlets.
2. GDELT file sizes are large and numerous, requiring reformatting via coding
in Python or another language. Transcripts from NexisUni do not need this
file management.
3. Another GDELT feature is Television Explorer,10 an easy-to-use search panel
accessing television news transcripts based on keyword searches. This can
help refine search terms for use in Nexis Uni. Although one cannot retrieve
texts from the Television News Explorer, the tool shows curves of term
frequencies over time to help decide on the time frame for analysis or to
segment time in organic intervals. Other useful features are sentiment mea-
surement (the tone of coverage), a word cloud, and thumbnails of the
videos and texts. These features can help verify that the search terms are
resulting in the desired outcomes.
90 J. A. Danowski, B. Yan, and K. Riopelle

4. Choosing the time interval for the analysis requires some exploration of the
series. Plotting the frequencies of term occurrence of terms may reveal nat-
ural segments that are not chronological but based on the distributions over
time. For example, over a year, one may observe a surge and subside over
several months, followed by recurrent mini surges over the next year. Perhaps
each of these surges becomes the organic segments for analysis. Otherwise, if
calendar time is used to segment texts, the choice of monthly, weekly, or
daily intervals depends on each text’s distribution. If there is sufficient text
volume, one can choose a daily interval, but we do not recommend this as a
starting slice. We find that it is better to begin with the larger interval, for
example, monthly, and decide whether this provides adequate variation
across the intervals. Then, we may move to a smaller slice, such as weekly or
daily. The natural limits on human processing are that if one relies on visual
interpretation and presentation of results in figures and tables, most
researchers could deal with up to a dozen intervals before overload occurs.
Otherwise, statistically, the daily interval is preferred.
5. For sentiment analysis, the SENET software we developed is target-specific, in
which the user selects a single word for which positive and negative senti-
ment scoring is desired. If the term is not a natural unigram but a string of
words, one can edit the text file accordingly to recode the string as a uni-
gram, for example, changing “social distancing” to “socialdistancing”.

Acknowledgments
The authors are grateful to George Barnett, Ke Jiang, and Aaron van Klyton for
reviewing the manuscript.

Notes
1 https://books.google.com/ngrams
2 This section is extracted from Danowski, Yan, and Riopelle (2020).
3 Smith et al. (2010), see also http://nodexl.codeplex.com from the Social Media
Research Foundation, http://www.smrfoundation.org
4 https://knightfoundation.org/reports/american-views-2020-trust-media-and-democracy/
5 https://mediabiasfactcheck.com/
6 https://www.allsides.com/unbiased-balanced-news
7 https://variety.com/2020/tv/news/network-ratings-2020-top-channels-fox-news-cnn-
msnbc-cbs-1234866801/
8 https://blog.gdeltproject.org/now-live-updating-expanded-a-new-dataset-for-exploring-
the-coronavirus-narrative-on-television-news/
9 https://www.lexisnexis.com/en-us/professional/academic/nexis-uni.page
10 https://blog.gdeltproject.org/now-live-updating-expanded-a-new-dataset-for-exploring-
the-coronavirus-narrative-on-television-news/
 Partisan ideology and market share growth 91

References
Arapakis, I., Lalmas, M., Cambazoglu, B. B., Marcos, M. C., & Jose, J. M. (2014). User
engagement in online News: Under the scope of sentiment, interest, affect, and gaze.
Journal of the Association for Information Science and Technology, 65(10), 1988–2005.
Asghar, M. Z., Khan, A., Ahmad, S., & Kundi, F. M. (2014). A review of feature extraction
in sentiment analysis. Journal of Basic and Applied Scientific Research, 4(3), 181–186.
Baccianella, S., Esuli, A., & Sebastiani, F. (2010, May). Sentiwordnet 3.0: an enhanced lexical resource
for sentiment analysis and opinion mining. Proceedings of the Seventh Conference on Interna-
tional Language Resources and Evaluation (LREC) (Vol. 10, No. 2010, pp. 2200–2204).
European Languages Resources Association (ELRA), Valletta, Malta .
Baughman, J. L. (2011). The rise and fall of partisan journalism. https://ethics.journalism.
wisc.edu/2011/04/20/the-fall-and-rise-of-partisan-journalism/.
Becher, M., Stegmueller, D., Brouard, S., & Kerrouche, E. (2020). Comparative experi-
mental evidence on compliance with social distancing during the COVID-19 pandemic.
doi:10.2139/ssrn.3652543.
Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of
communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 2008
(10), 10008.
Borgatti, S. P., Mehra, A., Brass, D. J., & Labianca, G. (2009). Network analysis in the social
sciences. Science, 323(5916), 892–895.
Brandes, U. (2001). A faster algorithm for betweenness centrality. Journal of Mathematical
Sociology, 25(2), 163–177.
Campbell, A. M. (2020). An increasing risk of family violence during the Covid-19 pan-
demic: Strengthening community collaborations to save lives. Forensic Science International:
Reports, 2, 100089.
Carley, K. (1993). Coding choices for textual analysis: A comparison of content analysis and
map analysis. Sociological Methodology, 75–126.
Christensen, S. R., Pilling, E. B., Eyring, J. B., Dickerson, G., Sloan, C. D., & Magnusson,
B. M. (2020). Political and personal reactions to COVID-19 during initial weeks of social
distancing in the United States. PloS One, 15(9), e0239693.
Clauset, A., Newman, M. E., & Moore, C. (2004). Finding community structure in very
large networks. Physical Review E, 70(6), 066111.
Corman, S. R., Kuhn, T., McPhee, R. D., & Dooley, K. J. (2002). Studying complex dis-
cursive systems. Centering resonance analysis of communication. Human Communication
Research, 28(2), 157–206.
D’Andrea, E., Ducange, P., Bechini, A., Renda, A., & Marcelloni, F. (2019). Monitoring
the public opinion about the vaccination topic from tweets analysis. Expert Systems with
Applications, 116, 209–226.
Danowski, J. A. (1982). Computer-mediated communication: A network-based content
analysis using a CBBS conference. Annals of the International Communication Association, 6
(1), 905–924.
Danowski, J. A. (1993). Network analysis of message content. Progress in communication Sci-
ences, 12, 198–221.
Danowski, J. A. (2013). WORDij Version 3.0: Semantic Network Analysis Software. Chicago,
IL: University of Illinois at Chicago.
Danowski, J. A., & Park, H. W. (2014). Arab spring effects on meanings for Islamist web
terms and on web hyperlink networks among Muslim-majority nations: a naturalistic field
experiment. Journal of Contemporary Eastern Asia, 13(2), 15–39.
92 J. A. Danowski, B. Yan, and K. Riopelle

Danowski, J. A., & Riopelle, K. (2019). Scaling constructs with semantic networks. Quality
and Quantity, 1–13. doi:10.1007/s11135-019-00879-5.
Danowski, J. A., Yan, B., & Riopelle, K. (2020). A semantic network approach to measur-
ing sentiment. Quality & Quantity, 1–35.
Formisano, R. P. (2008). For The People: American Populist Movements from the Revolution to the
1850s. Chapel Hill, NC: University of North Carolina Press.
Fuchs, C. (2011). Foundations of Critical Media and Information Studies (Vol. 52). New York:
Taylor & Francis.
Harper, C. A., & Rhodes, D. (2020). Ideological responses to the breaking of COVID-19 social
distancing recommendations. https://scholar.google.es/citations?view_op=view_citation&
hl=en&user=fYqfyJIAAAAJ&citation_for_view=fYqfyJIAAAAJ:YOwf2qJgpHMC
Hart, P. S., Chinn, S., & Soroka, S. (2020). Politicization and polarization in COVID-19
news coverage. Science Communication, 42(5), 679–697.
Hogan, M. J., & Hogan, M. J. (1987). The Marshall Plan: America, Britain and the Reconstruc-
tion of Western Europe, 1947–1952. Cambridge: Cambridge University Press.
Lipsitz, K., & Pop-Eleches, G. (2020). The partisan divide in social distancing. doi:10.2139/
ssrn.3595695.
Jiang, K., Barnett, G. A., & Taylor, L. D. (2016). Dynamics of culture frames in international
news coverage: A semantic network analysis. International Journal of Communication, 10, 27.
Jurkowitz, M., Mitchell, A., Shearer, & Walker, M. (2020). US media polarization and the
2020 election: A nation divided. https://www.journalism.org/2020/01/24/u-s-media
-polarization-and-the-2020-election-a-nation-divided/.
Kharde, V., & Sonawane, P. (2016). Sentiment analysis of twitter data: A survey of techni-
ques. arXiv preprint, 1601, 06971.
Khoo, C. S., & Johnkhan, S. B. (2018). Lexicon-based sentiment analysis: Comparative
evaluation of six sentiment lexicons. Journal of Information Science, 44(4), 491–511.
Kumar, S., & Zymbler, M. (2019). A machine learning approach to analyze customer satis-
faction from airline tweets. Journal of Big Data, 6(1), 1–16.
Kundi, F. M., Ahmad, S., Khan, A., & Asghar, M. Z. (2014). Detection and scoring of internet
slangs for sentiment analysis using SentiWordNet. Life Science Journal, 11(9), 66–72.
Liu, B. (2012). Sentiment analysis and opinion mining. Synthesis Lectures on human Language
Technologies, 5(1), 1–167.
Liu, B., & Zhang, L. (2012). A survey of opinion mining and sentiment analysis. In Mining
Text Data (pp. 415–463). Boston, MA: Springer.
Mäntylä, M. V., Graziotin, D., & Kuutila, M. (2018). The evolution of sentiment analysis—A
review of research topics, venues, and top cited papers. Computer Science Review, 27, 16–32.
Mikolov, T., Sutskever, I., Chen, K., Corrado, G., & Dean, J. (2013). Distributed repre-
sentations of words and phrases and their compositionality. arXiv preprint, 1310, 4546.
Monge, P. R., & Contractor, N. S. (2003). Theories of Communication Networks. New York:
Oxford University Press.
Osgood, C. E., Suci, G. J., & Tannenbaum, P. H. (1957). The Measurement of Meaning (No. 47).
Champaign, IL: University of Illinois Press.
Panchal, N., Kamal, R., Orgera, K., Cox, C., Garfield, R., Hamel, L., & Chidambaram, P.
(2020). The Implications of COVID-19 for Mental Health and Substance Use. San Francisco,
CA: Kaiser Family Foundation.
Pontiki, M., Galanis, D., Papageorgiou, H., Androutsopoulos, I., Manandhar, S., Al-Smadi,
M., & Eryiğit, G. (2016, January). Semeval-2016 task 5: Aspect based sentiment analysis.
In International Workshop on Semantic Evaluation (pp. 19–30). Beijing: Association for
Computational Linguistics.
 Partisan ideology and market share growth 93

Ramos, G., Vieites, Y., Jacob, J., & Andrade, E. B. (2020). Political orientation and support
for social distancing during the COVID-19 pandemic: evidence from Brazil. Revista de
Administração Pública, 54(4), 697–713.
Rogers, E. M. (1987). Progress, problems and prospects for network research: Investigating
relationships in the age of electronic communication technologies. Social Networks, 9(4),
285–310.
Rothgerber, H., Wilson, T., Whaley, D., Rosenfeld, D. L., Humphrey, M., Moore, A., &
Bihl, A. (2020). Politicizing the covid-19 pandemic: Ideological differences in adherence
to social distancing.
Rudkowsky, E., Haselmayer, M., Wastian, M., Jenny, M., Emrich, Š., & Sedlmair, M.
(2018). More than bags of words: Sentiment analysis with word embeddings. Commu-
nication Methods and Measures, 12(2–3),140–157.
Schiller, H. I. (1991). Culture, Inc: The Corporate Takeover of Public Expression. Oxford:
Oxford University Press on Demand.
Schiller, H. (2013). Information Inequality. London: Routledge.
Schneider, M. (2020, December 28). Year in review: Most-watched television networks —
ranking 2020’s winners and losers, Variety. https://variety.com/2020/tv/news/network-ra
tings-2020-top-channels-fox-news-cnn-msnbc-cbs-1234866801/.
Schwebel, A. I., & Cherlin, D. L. (1972). Physical and social distancing in teacher-pupil
relationships. Journal of Educational Psychology, 63(6), 543.
Simmons, S. J. (1978). The Fairness Doctrine and the Media. Berkley, CA: University of Cali-
fornia Press.
Smith, M., Ceni A., Milic-Frayling, N., Shneiderman, B., Mendes Rodrigues, E., Leskovec,
J., Dunne, C. (2010). NodeXL: a free and open network overview, discovery, and
exploration add-in for Excel 2007/2010/2013/2016. http://nodexl.codeplex.com/ from
the Social Media Research Foundation, http://www.smrfoundation.org.
Tausczik, Y. R., & Pennebaker, J. W. (2010). The psychological meaning of words: LIWC
and computerized text analysis methods. Journal of Language and Social Psychology, 29(1),
24–54.
Thakur, V., & Jain, A. (2020). COVID 2019-suicides: A global psychological pandemic.
Brain, Behavior, and Immunity, 88, 952–953.
Thet, T. T., Na, J. C., & Khoo, C. S. (2010). Aspect-based sentiment analysis of movie
reviews on discussion boards. Journal of Information Science, 36(6), 823–848.
Wang, B., & Liu, M. (2015). Deep learning for aspect-based sentiment analysis. Stanford
University Report. https://cs224d.stanford.edu/reports/WangBo.pdf.
Wemple (2020, April 1). Opinion: CNN, MSNBC refused to carry full Trump coronavirus
briefing. Yay!https://www.washingtonpost.com/opinions/2020/04/01/cnn-msnbc-refu
sed-carry-full-trump-coronavirus-briefing-yay/.
Zhang, L., Wang, S., & Liu, B. (2018). Deep learning for sentiment analysis: A survey. Wiley
Interdisciplinary Reviews: Data Mining and Knowledge Discovery, 8(4), e1253.
5
POLITICIZING THE HOLOCAUST
A comparative analysis of Israeli and German
speeches

Maya Hadar, Regula Miesch and Elad Segev

Introduction
The Holocaust plays a major role in the politics and national identity of Israel
and Germany (Meyers & Zandberg, 2002). While 76 years have passed since
the end of the Second World War, and 73 years after the foundation of the
state of Israel, the Holocaust is still a key event in Jewish, Israeli, and German
history. Unsurprisingly, the memory of the Holocaust is repeatedly raised in
various public discourses that extend beyond commemoration, such as in poli-
tical discourses.
The special role of political rhetoric in speeches given by heads of state within
political discourse is widely accepted among scholars. Speeches delivered in official
Holocaust commemoration ceremonies do not typically include deliberative state-
ments; however, and in line with Thomas Mann’s (2009) perception that “everything
is politics”, such speeches do not solely relate to Holocaust memory and com-
memoration. Political leaders utilize this opportunity to rhetorically construct and
maintain distinct social and political narratives. As such, Holocaust-related speeches
constitute valuable material for analysis, such as the current study, aimed to provide
insights into the main themes and narratives prevalent in contemporary Holocaust-
related discourse in Israel and Germany.
The assumption underlying this book chapter is that the Holocaust is one of the
enduring aspects of German and Israeli collective memory. As such, both societies
are motivated to maintain a link to the Holocaust despite changing circumstances
(Brug, 1998). While acknowledging that collective memories are socially con-
structed, historically situated, and continually shaped by agents of memory, such as
the state and the political elite, our goal is two-fold: First, we aim to unravel the
topics and map the narratives comprising Holocaust-related discourse as depicted
by Israeli and German political leadership (Prime Minister Benjamin Netanyahu

DOI: 10.4324/9781003120100-6
 Politicizing the Holocaust 95

and Chancellor Angela Merkel). Second, we seek to explore both discourses and
compare them to find how the Holocaust is instrumentalized in contemporary
political discourse.
Simply put, this study examines the way Israeli and German heads of state use
the stage given to them for Holocaust commemoration to rhetorically construct
and maintain non-Holocaust-related narratives such as national identity or defining
the standing of their countries in world politics. Ultimately, the research seeks to
uncover and compare the extent to which the Holocaust and the memory thereof
are instrumentalized for political purposes in both countries.
Scholarship concerning political discourse and rhetoric, politics of memory in
general, and Holocaust memory in particular form the theoretical basis of this
research. We employ semantic network analysis to identify and extract the main
themes from a corpus containing a set of political speeches.

Rhetoric and political leaders

The examination of rhetoric used by political leaders mainly refers to the agenda-
setting ability of the political elite in general, and of state leaders in particular. Dis-
cussing presidential rhetoric, Campbell & Jamieson (1990: 5) argued that “Presidents
have the opportunity to persuade us to conceive of ourselves in ways compatible
with their views of government and the world”. Similarly, Cohen (1995, 2015)
identified American Presidents’ leadership role and the public’s receptivity as the
central factors determining presidential rhetoric’s influence. Likewise, drawing upon
eight case studies of US Presidents, Zarefsky (2004) acknowledged the significance of
rhetoric’s power in altering public opinion by defining key political matters. Inter-
esting findings are also produced by DiMaggio (2015), who demonstrated that media
coverage of presidential rhetoric is rarely critical, but rather supportive and typically
reinforces the message, thereby increasing its influence on the public. Given the
importance of political discourse and presidential rhetoric in shaping public opinion
and as speeches constitute the most essential form of presidential rhetoric (Hodges &
Nilep, 2007) they are worthy of further research.
We use semantic network analysis to study the discourse of the political elite.
Whereas such discourse may include speeches, interviews, and public addresses
given by senior members of the government, we chose to explicitly focus on
speeches delivered by the heads of state (the Israeli Prime Minister and the German
Chancellor), who often represent the official discourse (cf. Jackson, 2005: 26), and
are prominent agenda-setters. Still, it is important to note that the political elite is
neither the single source of political discourse (civil society, the media, popular
culture, etc.) nor does it fully control it (cf. Spencer, 2010).

Politicization of the Holocaust in Israel

The politicization of various aspects of life, a tradition inherited from pre-state of
Israel was not only extensive and far-reaching after nationhood but also meant to
96 M. Hadar, R. Miesch, and E. Segev

allow politics to enter areas “that are not necessarily political in other societies”
(Horowitz & Lissak, 1990: 198–199).
Nili (2011: 39) argued that “the living influence of the Holocaust as a fixed
analogy in Israeli collective consciousness … has been pervasive throughout Israel’s
history”. Often derived from collective traumas, fixed analogies pertain to the
representation of new issues using the same memory that carries a certain psycho-
logical construct. In this way, collective traumas from the past continue to influ-
ence the present (Nytagodien & Neal, 2004), notwithstanding the role of political
actors and powerful social groups in sustaining this effect (Wood, 1999).
Through the first years of the state of Israel’s existence, the way Israeli society
conceived and remembered the Holocaust mirrored the tendency to view issues
through a political prism (Weitz, 1995). Official voices dominated Israel’s Holocaust-
related public discourse1 and utilized the Holocaust as a political and educational tool
(Weitz, 1993; Zertal, 1998). The dynamics of Holocaust commemoration in Israel’s
first decades echoed a conflict of cultures and ideologies, in what was referred to as a
“functional dynamic of memory” in the state of Israel (Baumel, 1995; Yerushalmi,
1982: 22). Throughout the 1950s, the Holocaust was one of the central sources of
political strife between the National Religious Party and other Israeli political groups
(Bar, 2005); the former sought to attach a religious interpretation to the events of the
Holocaust (Baumel, 1988) and the latter were interested in linking the Holocaust to
the establishment of the state of Israel (Stauber, 2000).
The Holocaust was incorporated into Israel’s fierce political debates concerning
the question of reparations payments from the German Federal Republic. In 1952,
Ben-Gurion’s agreement with West Germany for reparations for Jewish property,
confiscated by the Nazis, alongside personal reparations for Holocaust survivors,
caused an unprecedented public outcry in Israel. It was the first time in which the
tendency for the politicization of Holocaust memory was revealed (Shapira, 1998).
The reparations agreement was instrumentalized by those who opposed the ruling
Labor party. Originating from across the political spectrum, the opposition to the
agreement leveraged the memory of the Holocaust in their rhetoric. This issue
occupied Israeli media and public opinion for months (Shapira, 1998; Weitz, 1995).
Holocaust rhetoric appeared in Ben-Gurion and other Israeli leaders’ speeches even
prior to the founding of the state, e.g., Ben-Gurion’s speech at the Zionist World
Conference in London on August 2, 1945. The Holocaust was used as evidence to
prove the justice of the Zionist concept, galvanized American Jewry into a political
force that played a central role in the struggle for the Jewish state, and possibly swayed
members of the United Nations who voted in favor of Israel’s establishment (Shapira,
1998). As the need to end the homelessness of the Jewish people was the focus of the
political debate for the establishment of a Jewish state, the Holocaust constitutes one of
the sources of legitimacy for the state of Israel—both internally and externally (Shapira,
1998).2 Consequently, Israel has been widely acknowledged as the spiritual and prac-
tical heir to all Holocaust victims (Baumel, 1995).
The psychological presence of the Holocaust, ingrained in Israeli collective
memory, was not abandoned even despite military victories, diplomatic successes, and
 Politicizing the Holocaust 97

peace agreements Israel signed with neighbors and former belligerents like
Egypt and Jordan. Such presence is typically illustrated in genuine fear of a
national catastrophe, genocidal threat perception, and the framing of external
threats in Holocaust-laden terms (Nili, 2011).3 Manifestations of the psycholo-
gical presence of the Holocaust can be found both within the Israeli public4
and the Israeli political leadership.
Holocaust-driven Israeli sense of threat and discourse is also evident in the
context of the Israeli Arab conflict. Prime Minister Begin justified the Israeli
invasion of Lebanon (1982) by comparing the Palestine Liberation Organization
to the Nazis and added that besieging Arafat in Beirut felt as though he were
going after “Hitler in his bunker” (Friedman, 1990). The second Intifada, a
period of heightened terrorism, was perceived by many in Israel as “a challenge
to the very fabric of its society” (Freilich, 2006: 636). Palestinian threats, which
cannot objectively be conceived as existential due to the disparity in military
capabilities, were nevertheless perceived and framed as such. Consequently, the
Holocaust played a significant role in explaining Israel’s position in the Arab-
Israeli/Palestinian conflict and contributed to the perception that threats against
Jews should not be underestimated (Shapira, 1998).
The Holocaust also stands at the heart of Israel’s security policy and heavily
affects its formation (Freilich, 2006). The psychological presence of the Holocaust,
and Nilli’s (2011) fixed Holocaust analogy also govern the discourse pertaining to
what Israel perceives as its greatest contemporary security threat—the Iranian
nuclear project and sponsorship of militant Muslim groups.
The official Israeli approach to the Iranian nuclear project is mainly derived from
Iran’s belligerent and potentially genocidal posture toward Israel. Fierce anti-Israeli,
anti-Zionist, and anti-Semitic5 rhetoric repeatedly voiced by Iranian President
Mahmoud Ahmadinejad plays into the all too familiar narrative of enemies of the
Jews, be them Nazis, Palestinians, or Iranians, who seek the obliteration of the
Jewish state. Consequently, it is no wonder that the public discourse of the Iranian
nuclear project includes Holocaust-related rhetoric. Across the political spectrum,
Israeli leaders portrayed Iran as an existential threat and often referenced the
Holocaust in their condemnation of Ahmadinejad’s radical regime.
Yet, no Israeli leader has ever framed the Iranian nuclear threat in Holocaust terms
to the extent of Binyamin Netanyahu, who served for many years as Prime Minister.
On many occasions, Netanyahu drew parallels between Iran and Nazi Germany and
between Europe’s appeasement of Hitler and the Iranian nuclear agreement, formally
known as the Joint Comprehensive Plan of Action (2015). A partial list includes his
speech in the annual United Jewish Communities General Assembly,6 his 2009 speech
to the UN General Assembly, his speech on International Holocaust Remembrance
Day (2013) in front of the Israeli cabinet, and his remarks at the Saban Forum in 2017.
The many references made by Israeli political leadership to the Holocaust or to a
“Second Holocaust” as stated above, express the entrenchment of the collective sense
of existential threat, alongside the prevalence of the Holocaust in contemporary poli-
tical discourse.
98 M. Hadar, R. Miesch, and E. Segev

The Holocaust in German politics

Viewed either as the standard for evaluating German political activity, as an una-
voidable burden, even as an imposing taboo, Germany has long been dominated
by the memory of the Holocaust and the Nazi past. The idea that the “past won’t
go away” (Langenbacher, 2003) is continuously present in German politics, both
domestic and international, since as early as the outset of the post-war period.
Exploring the way Germany’s past and present shape each other, Olick and Levy
(1997) argue that “the Holocaust exerts its power in German politics, both
mythically and rationally” (p. 925). They set that the dominant interpretation of
the Holocaust heavily constrains German political culture. This is evident by the
rhetoric used by German leaders and in what Olick and Levy referred to as the
“ritualized politics of regret” (p. 933).
In its early days, West German leaders emphasized the contrast between the
Nazi past and the democratic present to regain the legitimacy of both their gov-
ernment and the German people (Rabinbach, 1988). Consequently, a central fea-
ture of the post-war West German political culture was its official anti-Nazi stance.
Amending provisions that allowed fragmentation at the political center (among
other things), the Basic Law of the new Federal Republic (September 1949) con-
stituted both a significant manifestation of this general feature and a “rejection of
the past” (Seifert, 1989: 40).
The rhetoric used by Chancellor Konrad Adenauer, among other early
German political leaders, served to support a metaphoric buffer created between
past and present (Greiffenhagen & Greiffenhagen, 1993). Such rhetoric high-
lighted, for example, that the new constitutional provisions “rectified the pro-
blems that had allowed Germany to be ‘seduced’ by bands of criminals” (Olick
& Levy, 1997: 926). Commitment to “Western” values and reparations to Israel
further supported this notion, protected the new Germany from its past problems, and
remedied the burdens of the Nazi period (Olick & Levy, 1997: 926, 927). The
rhetoric used by the West German political elite in the 1950s expressed this
notion—when addressing the Nazi past, political leaders clearly and repeatedly
stated that this was no longer a concern for Germany.7 Against the backdrop of
a wave of anti-Semitic vandalism and attacks on a synagogue in Cologne
(1958–1959), Adenauer dismissed accusations that significant anti-Semitism
exists in Germany.8
Anti-Semitism had been “solved” with reparations, and an official philo-Semitism
and pro-Israelism expressed by German politicians from across the political spec-
trum. Whereas support of Israel from the political left significantly diminished after
the Six-Day War (1967), West Germany continued to exhibit a positive and
uncritical attitude toward Israel regardless of its military or political actions (Zuck-
ermann, 1999; cf. Wolffsohn, 1988). Accordingly, Germany’s “special” relationship
with Israel (Deutschkron, 1991) has not only been part of West Germany’s politics
(Wolffsohn, 1988) but also “set the rules of the game for memory and culture for
the next half-century” (Olick & Levy, 1997: 928).
 Politicizing the Holocaust 99

Normalization is a central element of the German discourse around the

Holocaust (Zuckermann, 1999). The German desire to normalize the past has
been a characterizing feature of cultural politics in West Germany at least since
the mid-1980s (Olick & Levy, 1997). Chancellor Helmut Schmidt’s comment
made in Saudi Arabia in 1981 that German foreign policy should no longer be
“held hostage” to Auschwitz illustrates this notion (Wolffsohn, 1988: 42).
Throughout the 1950s, Israel repeatedly refused West Germany’s attempts to
establish diplomatic relations. When Israel finally agreed to establish official rela-
tions in 1965, German leaders justified the event in moral terms: Germany’s moral
duty to support Israel against Arab countries. In contrast, the realpolitik circum-
stances in the background (East German leader Walter Ulbricht’s state visit to
Egypt) were understated (Olick & Levy, 1997).
In the mid-1980s the Holocaust-centered memory was challenged by another
memory based on the same period of history, emphasizing German suffering. Whereas
the latter was not new, political advocacy of the German-centered memory re-
emerged when the center-right returned to power in 1982 (Maier, 1988).9 The “His-
torians’ Debate” (Historikerstreit) in the late 1980s’ constituted the central battle of
memory. Several prominent academics debated on memory regime dominance (Lan-
genbacher, 2003). The discussion mainly revolved around the status of the Nazi past in
German history and its implications for contemporary German identity (Holocaust
memory vs. forgetting). It also constituted an arena for a direct clash between the
Holocaust-centered memory and the German-centered memory. As such, the debate
was not “on more or less memory, but on the politics of memory” (Bartov, 2001: 298).
In a speech delivered to German historians in 1988, ending the “Historian’s
Debate” German President Richard von Weizsäcker stated that Germany must face
its historical responsibilities (Bulletin of the Press and Information Bureau of the
Federal Government, #131: 1185–1188). Whereas most listeners supported von
Weizsäcker’s statement, a similar statement, made a decade prior by President
Walter Scheel, received little to no notice. This example indicates the changing
discourse in German society and politics, from distancing the past to normalizing it
(Olick & Levy, 1997). Commemorations held on the 50th anniversary of the end
of the war further supports this notion, as Helmut Kohl’s government seemed to
have stepped back, to some extent, from its contentious attempts to “displace the
Nazi period from German identity … The Federal Republic has achieved “nor-
malcy” with regard to its past, not without it” (Olick & Levy, 1997: 933).
The German-centered memory regime had been rigorously “delegitimized and
made politically and culturally unacceptable by the ‘critical’ advocates of the Holo-
caust-centered one” (Langenbacher, 2003: 58). Supporters of the Holocaust-centered
memory managed to convince people that, as stated by German politician Antje
Vollmer “Auschwitz left no room for the subject of German tribulations” (Finn,
2002). Another result of the hegemony of the Holocaust-centered memory is the
unconditional need to combat right extremism. For example, following a series of
xenophobic and anti-Semitic attacks in 2001, Chancellor Gerhard Schröder called for
an “uprising of the upright” (Langenbacher, 2003).
100 M. Hadar, R. Miesch, and E. Segev

Advocates of the Holocaust-centered memory managed to maintain its dom-

inance, both politically and culturally, even a decade after German unification
(1990). Yet, the German-centered memory continued to re-emerge, particularly
during the “commemoration cycle” of 1995 and the publication of Grass’s novel
“Im Krebsgang” in early 2002 (Langenbacher, 2003).10 As representation of
German-centered memories has been monopolized by the political right for many
years, Langenbacher posits that the “latest historical discussions and debates represent
the left coming to terms with the memory of German suffering, just as conservatives
came to terms with the memory of German crimes in the 1980s and early 1990s”
(2003: 62). That being said, the Holocaust-centered memory that has been culturally
hegemonic for over two decades and is supported by a variety of commemorative,
pedagogical, and official measures has not been dislodged (Langenbacher, 2003).
Both memories coexist in the contemporary German memory landscape.
Today Holocaust-centered memory appears to be part of the political discourse
in Germany and to exert a significant influence among both elites and average
citizens; Former Foreign Minister Joschka Fischer’s (1998–2005) support of a quick
expansion of the EU to the east as “duty” due to Nazi crimes committed there
constitutes another example of the permeation of the collective memory of the
Holocaust into contemporary German politics (Banchoff, 1999, p. 178). Topics
such as euthanasia and genetic experimentations are still far more problematic to
openly discuss in Germany than in other countries (Langenbacher, 2003). German
involvement in contemporary conflicts is also a sensitive topic due to the German
past. If Germany were to make a military or financial contribution (such as to the
international coalition against Iraq in the context of the Gulf War of 1991, or the
Syrian Civil War of 2011) it could be accused of hostility. In contrast, non-con-
tribution may be considered careless and irresponsible (Olick & Levy, 1997).
A similar tension appears when immigration and asylum policies are concerned;
Germany struggled to reconcile its commitment to the right of asylum and support
of immigrants, due to the Nazi past and the Holocaust, with the high costs of
economic refugees entering Germany in the late 1980s and 1990s (Olick & Levy,
1997). Merkel’s “Refugees Welcome” policy pertaining to Syrian asylum seekers
has culminated in the arrival of over a million refugees in Germany since 2015 and
further illustrates the tension between seemingly moral obligations imposed by the
past and political decisions made in the present.
According to König (2015), references to the Nazi period have diminished after
German reunification but not entirely disappeared from the German political dis-
course. König predicted that such references, in which the Nazi past is only nominally
the topic, will become more frequent and will be utilized to attract public attention.
In short, since the Second World War, the Holocaust has been dominant in the
political discourse and in the implementation of national, regional, and even global
policies in Israel and Germany. In his book Zweierlei Holocaust: der Holocaust in den
politischen Kulturen Israels und Deutschlands,11 Zuckermann indicates that both Ger-
many and Israel instrumentalize the commemoration of the Holocaust to achieve
political goals. While Germany has used the Holocaust to present itself as a purged
 Politicizing the Holocaust 101

and a “normal” nation, Israel did so to constitute the newly formed Israeli society
(Zuckermann, 1999). Our goal in this chapter is to demonstrate the use of semantic
networks to highlight and compare those different narratives in the current political
discourse.

Methods
Semantic network analysis has been increasingly utilized in the study of political
discourse in general (Baden, 2010; Jung & Park, 2015), on social media (Breslin et
al., 2020; Eddington, 2018; Liu et al., 2018; Park et al., 2019), and when focusing
on political speeches and rhetorical presidency, often in a comparative analysis
(Chung & Park, 2010; Ghazal Aswad, 2019; Kuehl, 2012; Shim et al., 2015). In
the current study, Semantic network analysis is utilized in order to identify central
themes and underlying concepts governing the political discourse pertaining to the
Holocaust. It is a comparative study of speeches given within an identical time-
frame (2009–2020) by Netanyahu, the Israeli Prime Minister, and Merkel, the
German Chancellor.

Sample
We searched for Netanyahu’s speeches that mention the word “Holocaust”
between 2009 and 2020 on the Prime Minister’s Office website and the Gov-
ernment Services and Information website. Out of about 500 search results, we
manually selected 51 public speeches in which the Holocaust was mentioned and
discussed. Similarly, we searched for Merkel’s speeches that mentioned the terms
“Holocaust”, “Shoah”, and “Shoa” between 2010 and 2020 on the Chancellor’s
office website (https://www.bundeskanzlerin.de/bkin-de/aktuelles). We manu-
ally selected 31 relevant public speeches in which the Holocaust was mentioned
and discussed.

Procedure
We generated a list of the most frequent words that appeared more than 0.02%
(out of 57,159 words in Netanyahu’s speeches and 47,866 words in Merkel’s
speeches). We then inspected each word and cleaned the list from stop-words and
words that were too ambiguous or general. After cleaning the list we ended up
with a total of 147 words in Hebrew (from Netanyahu’s speeches), and 164 words
in German (from Merkel’s speeches). For each list we calculated the frequency
each pair of words appeared together in the same sentences and thus constructed
semantic networks. Finally, we used Visone (Brandes & Wagner, 2004) to calculate
the degree and betweenness centrality of words as well as to identify the clusters
(based on Louvain modularity, see also Blondel et al., 2008 and a more detailed
explanation of this procedure in Chapter 1).
102 M. Hadar, R. Miesch, and E. Segev

Results
Figure 5.1 portrays the semantic network of Netanyahu’s speeches mentioning the
word “Holocaust”. The word “Israel” was removed from the analysis as it appeared
at the absolute center and obscured the overall structure of the network. The clean
network includes pairs of words that appeared together in at least seven sentences
(0.12% of the total number of sentences). Following the procedure described in
Chapter 1, we conducted cluster analysis using the Louvain method to identify the
reoccurring themes in those speeches.
As can be seen the network portrays well the pathos–logos–ethos rhetoric of
Netanyahu, which appears in four clusters. The first very large cluster (labeled as
“them” at the bottom-center) defines the problem. It is the pathos, or what
Netanyahu perceives as the main present enemy of Israel—Iran. Apart from “Iran”
it includes the words “nuclear”, “missiles”, “weapon”, “develop”, “war”, and
“terror”. These words have mostly negative connotations, aiming to trigger the
emotions of the audience. The semantic network reveals that the Holocaust nar-
rative as presented by Netanyahu focuses on the constantly pending threat by Iran.
The pathos provides a proper justification for the logos. The upper cluster in the
middle is related to the main Zionist narrative (labeled as “us” at the top-center). It
includes the very central words “state” and “Jewish”, as well as “Herzl”, “army”,
“strong”, and “protect”. It reflects the reasons for the establishment of Israel as a
Jewish state, as a direct result of the Holocaust. Its very existence as Netanyahu

FIGURE 5.1 Semantic network of Netanyahu’s speeches mentioning the Holocaust.

Note. The size of a node indicates its betweenness centrality. The width of the ties
indicates the number of sentences in the corpus that mentioned each pair of words.
Clusters of countries are based on Louvain modularity. Network visualization and cal-
culations were performed with Visone (Brandes & Wagner, 2004).
 Politicizing the Holocaust 103

claims depends on a strong army that can protect the Jewish people from their
enemies to never experience such trauma again.
In between the pathos and the logos, which consist of the main narrative of “us”
against “them”, Netanyahu’s speeches also include the ethos aspect of rhetoric
through two parallel narratives of “me” and “you”. The ego narrative (labeled as
“me” on the right) includes the very central word “I” surrounded by words such as
“believe”, “negotiation”, “hope”, “achieve”, “peace”, “security”, and “truth”.
Thus, conveying that he is the only figure that ensures peace and security for the
state of Israel and stressing his personal achievements. The word “believe” is par-
ticularly important in this cluster, as it helps Netanyahu to associate himself with
those achievements. The combination of words “I believe” frequently appears in
Netanyahu’s speeches, and in Hebrew it can also be translated as “I trust”. It is
often connected with a process of negotiations for peace in the region and the
stability of Israel as a Jewish state.
Another important word in this cluster is “Begin”, which refers to Menahem
Begin, one of the main characters that Netanyahu repeatedly mentions in his
speeches. A leader of a Zionist militant group in the 1940s and the founder of the
right-wing party “Likud”, Begin is portrayed as the ideological father in Netanya-
hu’s conservative narrative. The centrality of the word “Begin” in Netanyahu’s
speeches is no coincidence. Similar to Begin, Netanyahu grew up in a revisionist
family, with the ideology developed by Ze’ev Jabotinsky, who supported territorial
maximalism and the establishment of a strong independent army to protect the
borders of Israel. Much of Begin’s fierce ideology is rooted in the Holocaust, and
when first meeting President Carter, he used the word “tertiated” to describe how
one in three Jews were murdered (JTA, 1977). Begin therefore used the Holocaust
to obtain political goals, a strategy that Bibi repeats. At the same time Begin is also
known for his significant achievement as a Prime Minister when signing the first
peace treaty with Egypt in 1979. Interestingly, Netanyahu’s settlement policy in
the West Bank, on the one hand, and his recent attempts to normalize the rela-
tionships of Israel with the Arab world, on the other, reflect a very similar pattern.
The fourth cluster (labeled as “you” on the left) includes many of the speech
addressees with names of official actors such as “President”, “Holocaust survivors”,
and other community members to whom the speech is targeted.
In short, Netanyahu’s Holocaust-related speeches predominantly focus on the pre-
sent (rather than the past) threat, and his political ambitions to strengthen the Zionist
narrative. Interestingly, the Holocaust in its historic dimensions is marginalized in his
speeches and is mainly told through the political lenses of international relations and
the ongoing threat on the existence of Israel. Words such as “history”, “remember”,
or “lesson” were frequent but still did not connect to other words in the network.
This means that general historical references were mentioned in Netanyahu’s speeches,
but they were not coherent enough to be part of the main narrative that he attempted
to promote.
For the German Chancellor, Angela Merkel, we analyzed 31 speeches that men-
tioned the words “Holocaust”, “Shoa”, or “Shoah” for the time period between
104 M. Hadar, R. Miesch, and E. Segev

March 2010 and September 2020. Merkel talks about the Holocaust mainly during
memorial ceremonies for victims of the Nazi regime or when she is given awards for
her efforts to fight anti-Semitism and promote human rights. The German Chancel-
lor, however, does not speak at the official Holocaust Memorial Hour in the German
Parliament. These speeches are given by the President of the Parliament and were not
included in this study. Figure 5.2 portrays the semantic network of Merkel’s speeches
mentioning the Holocaust.
Unlike Benjamin Netanyahu, Angela Merkel focuses on the country speaking in
the name of “Germany” and not herself. This reflects the effort to convey neu-
trality and modesty in regard to the issue and avoid the impression of pursuing
personal interests. The clusters in Figure 5.2 offer a chronological overview of
Germany’s role over history: the dominating cluster in the middle and center of
the network refers to Germany’s role in the past, naming the crimes committed
during the period of National Socialism. Merkel evolves her speeches around the
historic facts and the responsibility of Germany for the past atrocities. The word
“people” is given a prominent place at the center of the cluster. Together with
words describing the events such as: “persecution”, “murdered”, “suffering”, and
“survivors”, this cluster presents the emotionality underlying her speeches. It can be
seen as the core pathos, or emotional, part of the discourse, from which the
remaining clusters (the logos, or the reason) are developed. These clusters represent
the present and the future that grow out of the historical events.

FIGURE 5.2 Semantic network of Merkel’s speeches mentioning the Holocaust.

Note. The size of a node indicates its betweenness centrality. The width of the ties
indicates the number of sentences in the corpus that mentioned each pair of words.
Clusters of countries are based on Louvain modularity. Network visualization and cal-
culations were performed with Visone (Brandes & Wagner, 2004).
 Politicizing the Holocaust 105

The cluster at the bottom represents the moral aspects with which Germany deals
until up to today: guilt and responsibility. The guilt of the Holocaust is the under-
lying driving force that shapes the present and future policies of Germany. Terms
like “consciousness”, “remember”, and “everlasting” stress that this historic guilt
constitutes an eternal and unchangeable part of the German collective memory.
“Responsibility” is the central term of the cluster which refers to the eternal
responsibility that Merkel imposes on the German people for the crimes committed
in the past, but also underlining her promise not to let anything similar happen again.
Her speeches can clearly be attributed to the Holocaust-centered memory.
The cluster on the top represents the relations with the Jewish people in Ger-
many and the German–Israeli relations. Merkel stresses the promise of the present
government to guarantee security to the Jewish people and the state of Israel. This
represents another irrevocable element in the cluster and can be seen as the moral
counterpart of the underlying guilt. It can be interpreted as the overshadowing
oath of “never again”. Israel is usually mentioned together with positive words in
her speeches like “relations”, “connection”, or “safety”. By using the adjective
“German-Israeli”, Merkel underlines the close ties between the countries. Her
statements are usually pro-Israeli, and any criticism of Israeli politics (e.g., settle-
ment building) seems to be off-limits. When referring to “Jews” or “Jewish” she
often refers to anti-Semitism in Germany today and her dedicated stance against it.
The cluster on the left represents Germany as it perceives its own role in the
present. Angela Merkel portrays her country as a major partner and guarantor of
democracy, stability, and peace in Europe and the world. This is consistent with
Merkel’s and her party’s strong pro-European Union position, her party claiming
to be “the German EU party”.12 She as well often mentions the importance to
continue to fight anti-Semitism, thus depicting herself as the “watchdog” against
hate crimes. In this part of her speeches she stresses the transition of Germany from
“evil” to “good”. She illustrates Germany as a purged state that takes a morally
superior role in the concert of nations.
Finally, on the right we see the cluster referring to Germany’s responsibility for
the future. Merkel concentrates on the duty to defend values core to the German
government, such as “justice”, “democracy”, “tolerance”, and “freedom”, which
occupies a dominant place. “Freedom” is the core value that Merkel puts at the
center when talking about the present and future Germany. The values in this
cluster vastly correspond to the liberal-conservative values of Merkel’s center-right
party, the Christian Democratic Union (Christlich Demokratische Union
Deutschlands, CDU)—the title of the party program being “Freedom and Secur-
ity” (Freiheit und Sicherheit).13 She equally promotes her immigration policy during
the last years for which she has been heavily criticized by her political opponents
and which is believed to be one reason for the strengthening of the right-wing
populist party, Alternative for Germany (Alternative für Deutschland, AfD).14 By
connecting her party values and immigration policies to the “holy cow” of the
Holocaust, Merkel justifies her policies by the untouchable historical and moral
duty. By doing so, any criticism may be seen as devoid of morality.
106 M. Hadar, R. Miesch, and E. Segev

Conclusion
Against the backdrop of a dual process of collective memory preservation and con-
struction, depicted in the analysis, we explored the instrumentalization of the Holocaust
in contemporary political discourses. We extracted and compared the political speeches
of the Israeli Prime, Benjamin Netanyahu, and the German Chancellor, Angela Merkel,
mentioning the Holocaust over the last decade between 2009 and 2020. Using semantic
network analysis we identified the main themes and narratives of each.
Our findings show that both leaders use the Holocaust narrative to promote their
core values and their party values, yet each one of them focuses on a totally different
story and rhetoric strategy. Holocaust-related discourse as manifested in Netanyahu’s
speeches includes four distinct clusters: “us” (the state of Israel) and “them” (the ene-
mies of the state, currently Iran), “me” (the leader and the peace advocate) and “you”
(the audience). Netanyahu’s speeches demonstrate well the different rhetoric elements:
the pathos (which is related to the emotions and particularly fear), the logos (which is
related to the reason for establishing a Jewish state with a strong army), and the ethos
(the credibility of a strong leader). His speeches predominantly focus on the present
(rather than the past) threat, and his political ambitions to strengthen the Zionist nar-
rative. The current Iranian threat alongside other enemies of the state and the aspira-
tion for peace are framed as Netanyahu’s personal achievement.
The recurring linkage between the Holocaust, an event that occurred over 70
years ago and current security threats and enemies made by Netanyahu indicates
not only the enduring dominance of the memory of the Holocaust in 21st-century
Israeli consciousness but also its instrumentalization in political discourse. Such
instrumentalization is not the sole creation of Netanyahu but follows a tradition
that stretches back to before the establishment of the Jewish state.
In line with Van Dijk (1997: 11) approach to the study of political discourse,15 this
research also points to the practical results of the incorporation of the Holocaust in
contemporary political debate; Israeli Holocaust-fueled sense of threat leads to the
perception that similar to Hitler, the Iranian leader is, using the words of Netanyahu,
a fanatic (Netanyahu’s Remarks at the Saban Forum, 2017), and thus will not be
deterred by classic realist solutions to a potential nuclear contest (Nili, 2011). This
discourse may lay the foundations, legitimation, and even domestic and international
support for future Israeli military actions against the Ayatollah’s nuclearizing regime.
The inclusion of contemporary politics and political messages (e.g., concerning
domestic politics and current challenges such as the Iranian nuclear threat) in speeches
given by Netanyahu on Israel’s Holocaust Memorial Day reflects the existence of “the
present in the past” (Meyers & Zandberg, 2002: 390): It enables the political leadership
to take advantage of one of the most enduring aspects of Israeli collective memory and
to further construct this memory within a process of retelling.
Unlike Netanyahu, Merkel keeps a very limited use of the ethos element and does
not emphasize her own role as much. She rather tends to talk about (and in the name
of) Germany and the German people. While she uses the historical events of the
Holocaust themselves as the starting point of her argument, she emphasizes the changes
 Politicizing the Holocaust 107

in Germany from the past to the present and future and its international responsibility and
role to promote freedom, democracy, and tolerance. By stressing feelings of guilt and
the resulting obligation for compensation she promotes her party’s and her own lib-
eral-conservative ideas, with a strong pro-European and pro-Israel standing. Merkel’s
discourse of the Holocaust reflects Germany’s responsibility and international role to
prevent human suffering when refugees are concerned. This narrative further con-
tributes to our understanding of some of the motives that stood in the backdrop of the
“refugees welcome” migration policy.
In short, while Netanyahu uses the Holocaust to tell a story about the present and
the regional threat to Israel, Merkel is also looking at the past and the future, stres-
sing the changes in Germany, not only in the region but also in the global sphere. It
is important to note, however, that Netanyahu finds himself in a more or less
constant election campaign since 2009. With six elections between 2009 and 2020,
one can clearly see that the Holocaust–Iran nexus is a major aspect of his election
campaigns. It is his strategy to constantly mention the pending threat to the state of
Israel and him being the only candidate to save “his people” from doom. Merkel,
on the other hand, was elected and re-elected in three regular and relatively
uncontested elections since 2009.
Through an analysis of the Holocaust-related political discourse in both Israel
and Germany, we have illustrated that the past continuously interacts with the
present and is instrumentalized to pursue political goals. Consequently, our analysis
does not only constitute a contribution to discourse studies, but also to the study of
politics of memory, namely the way the political elite uses history and collective
narratives in order to promote political goals and shape national identity.

Tips and lessons from this process

1. Semantic network analysis is particularly useful when examining a large

corpus or conducting a comparative analysis utilizing large corpora. In this
chapter, we created two networks of words, based on all the speeches
pertaining to the Holocaust, made by the German Chancellor, Angela
Merkel, and the Israel Prime Minister, Benjamin Netanyahu. The speeches
varied in size and were delivered in a time frame of about ten years.
2. Semantic network analysis helps to identify the overall narrative of the spee-
ches. It enables to quickly map the main topics expressed in the speeches
alongside important words or terms. As such, and unlike other forms of con-
tent analysis, semantic network analysis helps the researcher to uncover the
main assumptions and general points of view expressed by the speakers.
3. Semantic network analysis of speeches is based on the transcribed text. It
is important to consider, however, that speeches are much more than the
text. Another important aspect is the way they are delivered. Some may even
consider speeches, notably political speeches, to be a performance. Audience
attention is sustained by a range of means (gestures, inflection, volume, tone,
108 M. Hadar, R. Miesch, and E. Segev

suspense, eye contact, passion, enthusiasm, etc.), which semantic networks

do not mirror.
4. Semantic networks are straightforward and sometimes fail to capture the
tone or the various layers of meaning. For example, irony, which normally
signifies the opposite meaning, is typically used in speeches to provoke
humor or empathy. Irony may not always be detected in semantic network
analysis, as it only registers the “natural” meaning of the words and expres-
sions used in the speech. This is why qualitative analysis is sometimes crucial
to complete the picture.

Notes
1 The extension of politics into all aspects of life may be seen as a desire to accumulate political
power, alongside an expression of politics as a way of influencing society (Don-Yehiya,
1993; Weitz, 1995).
2 An expression of this attitude can be found in the fact that all visiting heads of state are
taken to visit the Holocaust museum, Yad Vashem, upon arrival in Israel.
3 For example, Prime Minister Begin’s decision to launch an air-strike against the Iraqi
nuclear reactor in Osirak (June 1981) was “a clear and equivocal consequence of the
Holocaust” (Naor, 2003; Ne’eman, 2003). This statement was confirmed by Aviezer
Ya’ari, Chief of the Research Division of the IDF’s Intelligence Branch during the
Osirak strike and by Begin himself, who stated that he would “not be the man in whose
time there will be a second Holocaust” (Ya’ari, 2003: 31).
4 Such as the sense of impending doom, felt by the Israeli public during the three weeks
prior to the Six-Day War (the “Waiting Period”), the post-1973 largescale protests
against Israel’s political and military leadership due to a public discourse dominated by
the feeling that “an irrevocable disaster was only marginally avoided” (Ben-Porat et al.,
2008; Nili, 2011: 40; Ram, 2000), and wide public support pertaining to the Israeli
nuclear program (Cohen, 1998: 237).
5 Ahmadinejad exclaimed that Zionists dominate many of the world’s centres of power,
wealth, and media (MEMRI, 2009).
6 In Netanyahu’s words: “It’s 1938 and Iran is Germany. And Iran is racing to arm itself
with atomic bombs … he [the Iranian president] is preparing for another Holocaust for
the Jewish state” (Hirschberg, 2006).
7 They posited that any other position would imply an acknowledgment of collective
guilt. The Israeli Prime Minister, Ben-Gurion, insisted on such acknowledgment against
the backdrop of the reparation agreement, yet Adenauer declined (Deutschkron, 1991).
8 Whereas this is an understandable tactic when past events pose constrains on present or future
actions of the policy, Olick and Levy suggest that “deeper issues involved with identity for-
mation and the problematics of self-understanding” were also at play (1997: 928).
9 For example, in 1985, the Helmut Kohl government passed a law that simultaneously
criminalized Holocaust denial and denial of the expulsion (Langenbacher, 2003).
10 According to the narrator in Grass’ novel: “Never should we have kept silent about all
that suffering simply because our own guilt was overpowering” (“Another Taboo
Broken”, Grass, 2002).
11 Translated as “Two kinds of Holocaust: The Holocaust in the political cultures of Israel
and Germany”.
12 Freiheit und Sicherheit. Grundsätze für Deutschland. https://www.cdu.de/system/tdf/m
edia/dokumente/080215-grundsatzprogramm-kurz_0.pdf?file=1&type=field_collection_
item&id=1919
 Politicizing the Holocaust 109

13 Freiheit und Sicherheit. Grundsätze für Deutschland. https://www.cdu.de/system/tdf/m

edia/dokumente/080215-grundsatzprogramm-kurz_0.pdf?file=1&type=field_collection_
item&id=1919
14 https://www.afd.de/migrationspolitik/
15 According to Van Dijk (1997: 12), problems in political science can be studied more
completely and adequately when it is realized that the issues have an important discursive
dimension.

References
Baden, C. (2010). Contextualizing frames in political discourse: Using semantic network analysis to
investigate political parties’ framing strategies in the Dutch EU referendum campaign. International
Communication Association Annual Conference, Singapore.
Banchoff, T. (1999). The German Problem Transformed: Institutions, Politics, and Foreign Policy,
1945–1995. Ann Arbor, MI: University of Michigan Press.
Bar, D. (2005). Holocaust Commemoration in Israel During the 1950s: The Holocaust
Cellar on Mount Zion. Jewish Social Studies, 16–38.
Bartov, O. (2001). Mirrors of Destruction: War, Genocide, and Modern Identity. Oxford: Oxford
University Press.
Baumel, J. T. (1995). “In everlasting memory”: Individual and communal Holocaust
commemoration in Israel. Israel Affairs, 1(3), 146–170.
Baumel, Y. T. (1988). ‘Zikhron olam,’tfilot vi-yemei evel le-ahar ha-Shoah le-zekher
korbenot ha-Shoah. Sinai, 101, 281–294.
Ben-Porat, G., Levi, Y., Mizruchi, S., Naor, A., & Tzfadia, E. (2008). Israel since 1980.
Cambridge: Cambridge University Press.
Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of com-
munities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 10, 10008.
Brandes, U., &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In M.
Jünger & P. Mutzel (Eds.), Graph Drawing Software (pp. 321–340). New York: Springer-Verlag.
Breslin, S. D., Enggaard, T. R., Blok, A., Gaardhus, T., & Pedersen, M. A. (2020). How we tweet
about coronavirus, and why: A computational anthropological mapping of political attention on Danish
Twitter during the COVID-19 pandemic. The COVID-19 Forum III. University of St. Andrews.
Brug, M. (1998). The vacillations surrounding Holocaust commemoration in Israel. In Y.
Rappel (Ed.), Memory and Awareness of the Holocaust in Israel (pp. 45–54). Tel Aviv: Israel’s
Defense Ministry Publications.
Campbell, K. K., & Jamieson, K. H. (1990). Deeds Done in Words: Presidential Rhetoric and the
Genres of Governance. Chicago, IL: University of Chicago Press.
Chung, C. J., & Park, H. W. (2010). Textual analysis of a political message: The inaugural
addresses of two Korean presidents. Social Science Information, 49(2), 215–239.
Cohen, A. (1998). Israel and the Bomb. New York: Columbia University Press.
Cohen, J. E. (1995). Presidential rhetoric and the public agenda. American Journal of Political
Science, 87–107.
Cohen, J. E. (2015). Presidential leadership of public opinion: An embedded survey
experiment. Political Communication, ahead-of-print, 1–11.
Deutschkron, I. (1991). Israel und die Deutschen: Das schwierige Verhältnis [Israel and the Ger-
mans: The Difficult Relationship]. Cologne: Verlag Wissenschaft and Politik.
DiMaggio, A. R. (2015). Selling war, selling hope: Presidential rhetoric, the news media, and US
foreign policy since 9/11. Albany, NY: SUNY Press.
Don-Yehiya, E. (1993). Memory and political culture: Israeli society and the Holocaust.
Studies in Contemporary Jewry, 9, 139–161.
110 M. Hadar, R. Miesch, and E. Segev

Eddington, S. M. (2018). The communicative constitution of hate organizations online: A

semantic network analysis of “Make America great again”. Social Media+ Society, 4(3).
doi:2056305118790763.
Finn, P. (2002, February 11). Debate is rekindled over WWII expellees. Washington Post. https://
www.washingtonpost.com/archive/politics/2002/02/11/debate-is-rekindled-over-wwii-exp
ellees/d2fbb97a-60b1-48d0-b98a-c17466da6451/.
Freilich, C. D. (2006). National security decision-making in Israel: Processes, pathologies,
and strengths. The Middle East Journal, 60(4), 635–663.
Friedman, T. L. (1990). From Beirut to Jerusalem. London: Macmillan.
Ghazal Aswad, N. (2019). Exploring charismatic leadership: A comparative analysis of the
rhetoric of Hillary Clinton and Donald Trump in the 2016 presidential election. Presidential
Studies Quarterly, 49(1), 56–74.
Grass, G. (2002, November 23). Another taboo broken. The Economist. https://www.
economist.com/europe/2002/11/21/another-taboo-broken.
Greiffenhagen, M., & Greiffenhagen, S. (1993). Ein schwieriges Vaterland: Zur politischen Kultur
im vereinigten Deutschland [A Difficult Fatherland: On Political Culture in United Germany].
Munich: List.
Hirschberg, P. (2006, November 13). Netanyahu: It’s 1938 and Iran is Germany; Ahmadi-
nejad is preparing another Holocaust—Haaretz Com—Haaretz.com. Haaretz. https://
www.haaretz.com/1.4931862.
Hodges, A., & Nilep, C. (2007). Discourse, war and terrorism (Vol. 24). Amsterdam: John
Benjamins Publishing.
Horowitz, D., & Lissak, M. (1990). Trouble in Utopia: The Overburdened Polity of Israel. Israel:
Am Oved.
Jackson, R. (2005). Writing the War on Terrorism: Language, Politics and Counter-terrorism.
Manchester: Manchester University Press.
JTA. (1977, August 11). Begin: No U.S. pressure on Israel. Jewish Telegraphic Agency. https://
www.jta.org/1977/08/11/archive/begin-no-u-s-pressure-on-israel.
Jung, K., & Park, H. W. (2015). A semantic (triz) network analysis of South Korea’s “open
public data” policy. Government Information Quarterly, 32(3), 353–358.
König, H. (2015). Die Zukunft der Vergangenheit: Der Nationalsozialismus im politischen
Bewußtsein der Bundesrepublik [The future of the past: National Socialism in the political con-
sciousness of the Federal Republic]. Frankfurt: S. Fischer Verlag.
Kuehl, R. A. (2012). The rhetorical presidency and “accountability” in education reform:
Comparing the presidential rhetoric of Ronald Reagan and George W. Bush. Southern
Communication Journal, 77(4), 329–348.
Langenbacher, E. (2003). Changing Memory Regimes in Contemporary Germany? German
Politics & Society, 21(67), 46–68.
Liu, W., Lai, C.-H., & Xu, W. W. (2018). Tweeting about emergency: A semantic network
analysis of government organizations’ social media messaging during hurricane Harvey.
Public Relations Review, 44(5), 807–819.
Maier, C. S. (1988). The Unmasterable Past: History, Holocaust, and German National Identity.
Harvard University Press.
Mann, T. (2009). Der Zauberberg: Roman. Frankfurt: S. Fischer Verlag.
MEMRI. (2009, April 2). Extracts from weekly speeches of Iranian President Mahmoud
Ahmadinejad. MEMRI. https://www.memri.org/reports/speeches-week-iranian-presi
dent-mahmoud-ahmadinejad-discusses-holocaust-saying-reexamining.
Meyers, O., & Zandberg, E. (2002). The sound-track of memory: Ashes and Dust and the
commemoration of the Holocaust in Israeli popular culture. Media, Culture & Society, 24
(3), 389–408.
 Politicizing the Holocaust 111

Naor, A. (2003). Analysis of the decision-making process. In Israel’s Strike Against the Iraqi
Nuclear Reactor. Jerusalem: Begin Center.
Ne’eman, Y. (2003). My involvement in the Osirak affair. In Israel’s Strike Against the Iraqi
Nuclear Reactor. Jerusalem: Begin Center.
Nili, S. (2011). The nuclear (and the) Holocaust: Israel, Iran, and the shadows of Auschwitz.
Journal of Strategic Security, 4(1), 37–56.
Nytagodien, R., & Neal, A. (2004). Collective trauma, apologies, and the politics of
memory. Journal of Human Rights, 3(4), 465–475.
Olick, J. K., & Levy, D. (1997). Collective memory and cultural constraint: Holocaust myth
and rationality in German politics. American Sociological Review, 921–936.
Park, S.-H., Yoon, S.-W., & Chung, S. W. (2019). Social and semantic network analysis of
Facebook pages: The case of Brexit and Bremain leaders. Asia Europe Journal, 1–14.
Netanyahu’s Remarks at the Saban Forum. (2017, March 12). Israel Ministry of Foreign Affairs.
https://mfa.gov.il/MFA/PressRoom/2017/Pages/PM-Netanyahus-remarks-at-the-Saba
n-Forum-3-December-2017.aspx.
Rabinbach, A. (1988). The Jewish question in the German question. New German Critique,
44, 159–192.
Ram, U. (2000). National, ethnic or civic? Contesting paradigms of memory, identity and
culture in Israel. Studies in Philosophy and Education, 19(5–6),405–422.
Seifert, J. (1989). Die Verfassung (The Constitution). In W. Benz (Ed.), Die Geschichte der
Bundesrepublik Deutschland, (Vol. 1, pp. 40–70). Frankfurt: S. Fischer Verlag.
Shapira, A. (1998). The Holocaust: Private memories, public memory. Jewish Social Studies, 4
(2), 40–58.
Shim, J., Park, C., & Wilding, M. (2015). Identifying policy frames through semantic network
analysis: An examination of nuclear energy policy across six countries. Policy Sciences, 48(1),
51–83.
Spencer, A. (2010). The tabloid terrorist: The predicative construction of new terrorism in the media.
New York: Springer.
Stauber, R. (2000). Ha-lekah la-dor: Shoah u-gvurah ba-mahashavah ha- tsiburit ba-arets bi-shnot
ha- hamishim. Jerusalem: Yad Ven Zvi.
Van Dijk, T. A. (1997). What is political discourse analysis? Belgian Journal of Linguistics, 11
(1), 11–52.
Weitz, Y. (1995). Political dimensions of Holocaust memory in Israel during the 1950s. Israel
Affairs, 1(3), 129–145.
Weitz, Y. (1993). The Shaping of Holocaust Memory in Israeli Society During the 1950s’.
In Y. Gutman (Ed.), Changes in the Jewish People After the Holocaust (pp. 473–493). Jer-
usalem: Yad Vashem.
Wolffsohn, M. (1988). Ewige Schuld? 40 Jahre Deutsch-Jüdisch-Israelische Beziehungen [Perpetual
Guilt? Forty Years of German- Jewish-Israeli Relations]. Munich: Piper.
Wood, N. (1999). Vectors of memory. Oxford: Berg.
Ya’ari, A. (2003). Intelligence aspects of the attack on the Iraqi nuclear reactor. In Israel’s
Strike Against the Iraqi Nuclear Reactor. Jerusalem: Begin Center.
Yerushalmi, Y. H. (1982). Zakhor: Jewish History and Jewish Memory.
Zarefsky, D. (2004). Presidential rhetoric and the power of definition. Presidential Studies
Quarterly, 34(3), 607–619.
Zertal, I. (1998). From Catastrophe to Power: The Holocaust Survivors and the Emergence of Israel.
Berkley, CA: University of California Press.
Zuckermann, M. (1999). Zweierlei Holocaust: Der Holocaust in den politischen Kulturen Israels
und Deutschlands. Munich: Wallstein Verlag.
6
NETWORK OF CLEAVAGES?
British paradiplomacy in the (digital) international
discourse around Brexit

Jérôme Chariatte and Diana Ingenhoff

Introduction
The decision of the United Kingdom to leave the European Union has shaken
Europe like an earthquake. In a world that is increasingly characterized by inter-
connectedness and globalization, an EU-member country for 47 years, decided to step
back and isolate itself from the European political arena. However, this development
does not come out of nowhere. Over the last few years, Euroscepticism has been
observed in many different European countries, resulting in the strengthening of the
so-called populist parties (Brack & Startin, 2015; FitzGibbon et al., 2017). Indeed, the
optimism that surrounded the rise of a “global society” starts to crack. According to
Flew (2020), one reason for the new interest in nationalist ideas is that citizens have
lost confidence in international politics and organizations, as they feel their domestic
needs are left behind. Indeed, international politics and public diplomacy sometimes
failed to consider the importance of domestic issues and national identity (Zaharna,
2020). It is crucial to take into account national identities as they influence interna-
tional politics and are an essential dimension of public diplomacy (Huijgh, 2019).
National identity is a complex phenomenon. Indeed, countries can show many
different identities and attitudes, as we can see in the British case: On June 23,
2016, 51.9% of the British people voted to leave the European Union (BBC,
2016a). This narrow gap can be explained by the diverse national identities in the
United Kingdom. In Northern Ireland, for example, a majority of 55.8% voted
to stay and in Scotland 62% of the electorates. As a consequence, Scotland con-
siders becoming an independent country. In England, in which 53.4% voted in
favor of dropping out, there was a “gap” between rural and urban areas. For
instance, across all boroughs of London, 59.9% voted against leaving the Eur-
opean Union. Regions and cities differ in their attitude toward international
politics and start to communicate their own agenda (BBC, 2016b).

DOI: 10.4324/9781003120100-7
 Network of cleavages? 113

The actions and interactions of sub-national entities, such as cities, regions or

sub-state actors, in international relations are often considered as a kind of parallel
diplomacy to a nation’s government, and is therefore defined as “paradiplomacy”
(Tavares, 2016). Yet the term “parallel” does not necessarily mean that their
activities are unrelated to the official government. They represent a bottom-up
force, which can influence the (inter)national policies of their governments (Kuz-
netsov, 2014). Despite its rising importance at the beginning of the 1990s with
publications from Michelmann and Soldatos (1990) and Aldecoa and Keating
(1999) and the first comparative studies on sub-state diplomacy in ten federal
countries (Schiavon, 2019), paradiplomatic communications still remain evidently
understudied (Duran, 2016).
The extent to which paradiplomatic voices shape the public international dis-
course depends heavily on their media presence. As described by Manuell Castells
(2011), in a “Network Society”, technology and new media platforms play a
decisive role in connecting people. Therefore, in order to understand the way
nations are perceived at home and abroad and to adapt communication strategies
accordingly, one must first understand the role played by the media in producing
national and international discourse.
Particularly relevant for governments and public diplomacy officials is to be
aware of sub-national actors and their interests in order to be able to coordinate
and adapt their (inter)national communications (Alvarez, 2020). Thus, public
diplomacy, as well as paradiplomacy, begins with listening, which means to under-
stand the different actor’s positions and opinions about a country or its related
issues and the properties of different communication channels (Cull, 2019; Di
Martino, 2019). Of the same opinion is MacDonald (2016: 284), who believes that
the future perception and reputation of the United Kingdom depends strongly on
how much attention it pays to the opinion of its relevant actors:

If the UK’s public diplomacy is to succeed, it will need to be smarter, parti-

cularly in its ability to describe and analyse multiple points of view, more
attuned not only to promoting the UK, and to the development of new
relationships.

In this chapter, we will employ semantic network analysis to analyze how the dis-
course around Brexit is shaped by sub-national actors. We hope to contribute to the
research on paradiplomacy and city diplomacy by illuminating it from a communica-
tion perspective. In what follows, we discuss the interplay between populism, national
identity, and different media platforms, and show the implications of issue framing on
country image formation and international communication strategies.

New voices in public diplomacy and the rise of paradiplomacy

Today’s diplomatic work is no longer limited to diplomats but needs to include all
kinds of publics in the international discourse. Traditionally, the term “public
114 J. Chariatte and D. Ingenhoff

diplomacy” describes the relationship management between one nation’s diplomats

and international audiences. However, in times where digital platforms allow
international audiences to interact with each other and for media information to
spread rapidly around the world, we can observe a shift “from the powerful state as
the main actor that has the control over this communication process toward other
actors and society that also contribute to and influence this process” (Dolea,
2018: 333). Thus, Melissen (2005: 22) introduces the notion of “New Public
Diplomacy”, where different kinds of actors, such as international organizations,
corporations, regions, or individuals, shape international relations, shifting from
one-way communication to a more dialogic and engagement approach.
A special interest should be given to the so-called sub-state actors: During the
whole Brexit process, Scotland claimed its European identity and protested against
the referendum to show its discontent. This reinforces Scotland’s efforts to position
itself as a nation with its own identity. For years, Scotland has exemplified how a
sub-state can engage in independent public diplomacy and branding activities to
cultivate its image (Criekemans, 2010). In the research field, the involvement of
representatives of non-central governments in international politics is largely
described under the term “paradiplomacy” (Alvarez, 2020).
Schiavon (2019) finds two big paradigms reviewing the literature on concepts,
models, and definitions. On the one hand, paradiplomacy can be seen as com-
plementary to the state’s foreign policy, and on the other these international activities
can jeopardize a unified and coherent national foreign policy. Kuznetsov (2014) points
out that there is a lack of consensus on the use of the term and what it contains. For
instance, scholars disagree on whether cities are to be considered as paradiplomatic
actors, and if they might challenge the integrity and power of a state.
We consider that cities do indeed play a major role, as they often represent the
economic and political hearts of countries. Throughout history, cities have played a
major role in the international arena, but it is especially in times of global trade and
increasing urbanization that cities could solidify their influence (Acuto et al., 2021;
Smith, 2019). Not only do cities represent major travel destinations when visiting a
country (Hanna & Rowley, 2019), they also strongly shape the national brand
(Dinnie & Sevin, 2020). In recent years, we can observe that cities engage inter-
nationally in climate issues, but also in human rights, migration, and health (Acuto
et al. 2017; Asdourian & Ingenhoff, 2020; Bäckstrand et al., 2017). By this, they
can often counter the politics of their nation-states, especially when related to
social or normative issues (Lievonen & Luoma-aho, 2015). As La Porte (2013)
highlights, a city’s influence lies primarily through the communication of ideas.
Thus, large cities, particularly metropolitan cities, are considered to take on the
role of “defenders of universal values” (Leffel, 2018: 504). These activities are often
considered as “city diplomacy”, which means “institutions and processes by which
cities engage in relations with actors on an international political stage with the aim
of representing themselves and their interests to one another” (Van der Pluijm &
Melissen, 2007: 6). For example, it would have been incomplete to analyze Pre-
sident Trump’s withdrawal of the United States from the Paris Agreement on
 Network of cleavages? 115

climate change without taking into consideration how state governors and mayors
endorsed the United States Climate Alliance through their paradiplomatic city
diplomacy activities (Asdourian & Ingenhoff, 2020).
In the context of Brexit, many British cities started information campaigns, as
they felt that their voices and interests were not considered and represented prop-
erly by the national government and, according to prognoses, they would feel
Brexit’s consequences particularly strongly (Szpak et al., 2020). The most recent
example is London’s handling of the Brexit issue. Shortly after the referendum,
London’s Mayor Sadiq Khan started his #LondonIsOpen campaign to reassure the
European citizens of London that they are still welcome, and to maintain rela-
tionships with international investors and visitors. Indeed, city diplomacy and city
branding campaigns are intended for domestic and international audiences as well
(Dinnie & Sevin, 2020; Duncombe, 2019).
Hence, the British capital holds on to its European identity and opposes itself to
the isolationist tendency of the British government. Starting with a concert of
Stevie Wonder in Hyde Park on July 10, 2016, the campaign was extended to
various communication channels, from posters in the London underground to
online advertising videos and social media postings. The success and unique char-
acter of the campaign was due to its collaborative orientation, as it specifically
invited social media users to co-create the message of London’s diversity and
openness on a large scale (Blay Arráez et al., 2019). It was not only related to the
volume of engagement but also to the long duration the campaign lasted. This was
due to the exceptional and explicit demand for co-creation, as previous studies
have shown that city campaigns on social media are rather one-way-directed,
short-termed, and focused on specific events (Sevin, 2016). Collaborative cam-
paigns can be particularly useful for the international communication of cities: In
early 2020, several mayors of European cities shared the slogan #LondonIsOpen in
their respective languages on Twitter to show their ongoing relationship with the
British capital.
Yet, in general, very little is known about how sub-national actors are being
perceived and discussed in the international sphere. In the context of Brexit, we
therefore intend to analyze (RQ1) How do paradiplomatic actors such as sub-states or
cities shape the Brexit discourse? In the next step, we would like to delve into the
extent to which digital media offerings enable (non-)political actors to shape
international communication, as co-creative campaigns might have great potential
for diplomacy.

Digital diplomacy and social media

Digital media challenges diplomatic communication and place branding: “The
global spread of new digital communication technologies has profoundly trans-
formed the way individuals, states and businesses operate and interact with the
outside world” (Bjola & Zaiotti, 2020: 1). Not only do social media allow to reach
big audiences with a small budget, which may be interesting for regional actors
116 J. Chariatte and D. Ingenhoff

such as cities (Mendes, 2013; O’Boyle, 2019), they also allow all kinds of users to
be involved in the communication and image building process (Sevin &
Ingenhoff, 2018). Therefore, the concept of digital diplomacy has gained a lot of
attention in recent years and changed the understanding of international relations
and diplomacy, meaning “a strategy of managing change through digital tools and
virtual collaboration” (Holmes, 2015: 15). Digital communication has also chan-
ged the patterns of international information flow, which can influence the
construction of the country image and place branding (Golan & Himelboim,
2016; Segev, 2018).
As highlighted by Manor and Segev (2015), the concepts of place branding and
digital diplomacy can overlap. Especially in times of crisis, digital diplomacy uses
branding strategies in order to mend or correct a country’s image after a crisis such
as Brexit for the United Kingdom. Thereby, promoting the country’s own identity
plays an important role, as nation branding can help a nation “to articulate a more
coherent and cohesive national identity, to animate the spirit of its citizens … and
maintain loyalty to the territory within its borders” (Aronczyk, 2013: 3).
In place branding literature, citizens and user-generated content in the communica-
tion process has become an important issue (Compte-Pujol et al., 2018; Dinnie & Sevin,
2020). This is mainly because citizens often do not feel that their identities are repre-
sented in international campaigns (Casais & Monteiro, 2019; Insch & Walters, 2018). At
the same time, the inclusion of citizens is essential, as they can give authenticity to
communication campaigns (Casais & Monteiro, 2019; Rickly-Boyd, 2012), a major
resource in times of fake news and institutional distrust. Further, by including more
actors in the campaign building, the campaign will gain more spread (Bjola et al., 2020).
One famous example was the Curators of Sweden project, which gave the official Twitter
channel @Sweden to citizens (Christensen, 2013). Despite this, current international
branding campaigns are still strongly based on a top-down approach and fail to make use
of the dialogic nature of social media sites (Kampf et al., 2015). Also, in the academic
literature, studies often analyze how governmental actors communicate on the social
web but do not consider the importance of non-governmental actors’ content in
country image formation in the digital sphere (Collins et al., 2019; Strauß et al., 2015).
Attention must also be given to the fact that in the digital sphere, both interna-
tional and national players are co-creating the image of a country (Melissen, 2011).
National and international actors have different identities and relationships with a
country, which can impact their communication patterns. For instance, Šimunjak
and Caliandro (2020) analyzed how and if the diplomats from the 27 European
countries reacted differently to the Brexit on Twitter. Results indicate that, even if
diplomats tried to communicate with neutral and consistent communication with the
other EU countries, there are some differences we can notice. Countries which are
traditionally British allies such as the Netherlands, Sweden, or Denmark were more
critical, whereas countries having big diasporas in Great Britain like France, Poland,
or Germany were more positive and focusing on the future bilateral relationships.
Finally, when analyzing digital content, it is also important to consider the spe-
cificities of the media channels (Andéhn et al., 2014). Country images on social
 Network of cleavages? 117

media sites are not only built through functional information, such as political or
economic news, but also strongly through cultural and natural aspects, such as the
country’s traditions and landscapes (Chariatte et al., 2019; Sevin & Ingenhoff,
2018). But more importantly, the discourse on social media sites such as Twitter or
Facebook is very immediate, dynamic, and short. In order to attract attention users
tend to keep their posts simple, provocative, and emotional. As proposed by
Duncombe (2019), we need to pay “greater attention to the interaction of social
media, emotion and identity, when analyzing today’s and future digital diplo-
macy”. Indeed, very few studies have analyzed the content shared on social net-
working sites and how it influences the country and place image and its related
issues (Ingenhoff et al., 2021). Our second research question would be therefore
(RQ2) How is Brexit discussed on Twitter?

The impact of international news on place image

Together with social media, international news is particularly important in
constructing and shaping the images of places and countries. Many studies have
shown that international news impacts the country image formation process
(Brewer et al., 2003; Wanta, 2018; Wanta et al., 2004). International news
represents one of the major information sources, especially about countries that
people do not or cannot easily have direct experiences with (Ingenhoff &
Buhmann, 2019; Zhang & Meadows III, 2012).
In its fundamental work “public opinion”, Lippmann (1946: 90) highlighted that
the media strongly define what we think about other countries, which is also true
for places: “We are told about the world before we see it. We imagine most things
before we experience them”. According to agenda-setting theory (McCombs &
Shaw, 1972), there are three layers on which international news can influence our
perceptions of countries. On the first agenda-setting level, media can impact the
salience of countries and thus on their importance by the number of news articles.
However, in order that a country is perceived as important for its own nation,
international news coverage needs to highlight the link to the home nation and its
policies (Wanta, 2018). Further, on a second and third level, international news has
an attribute agenda-setting effect. In a network agenda-setting approach and in line
with associative network theory (Andersen, 1983), it is assumed that people link
countries to a set of attributes that are displayed in the news (Wanta, 2018).
In the news media it can be observed that the discourse on the topic of Brexit
has changed and hardened over time. As shown by Bijsmans et al. (2018), inter-
national news in France, the Netherlands, and Germany even perceived the Brexit
negotiation as a positive impulse for a reformation of the European Union at the
beginning, but over time it has been perceived more and more as a danger that
must be countered by the European countries as a unit, in order to avoid similar
tendencies in Europe to grow.
Europe’s united and tough stance on Brexit has a significant impact on the UK’s
future international relations. By exploring the speeches of international high-level
118 J. Chariatte and D. Ingenhoff

decision-makers, Oppermann et al. (2020) conclude that, despite the British desire
for independence, the United Kingdom’s role is strongly dependent on the Eur-
opean Union. Already, at the beginning of the Brexit negotiations, many countries
made it clear that their relationship with the EU is more important to them in
economic and political affairs, and they did not want to jeopardize their relations
by negotiating new trade deals with the United Kingdom. The United Kingdom’s
desire to become even more global after Brexit by making new international trade
deals may result in an isolated position.
This dependency on external partners goes along with the assumption of role
theory (see Biddle, 1986; Harnisch et al., 2011). It states that the roles of actors arise
from their own ideas, but also strongly from the expectations of others. “Britain’s
foreign policy after Brexit will not be defined simply by what role Britain wants to
play, but equally by what role other states let Britain play” (Oppermann et al.,
2020: 134). This highlights the relevance of listening to these international voices
about Brexit, as they may have an impact on the United Kingdom’s future strategic
communication. Therefore, we examine (RQ3) How is Brexit depicted in interna-
tional mass media?

Methods
In the present study, we adopt a listening approach. This means that although we
disclose our theoretical assumptions and ideas regarding the different actors in the
international discourse, we would like to pay attention to and understand the
diversity of opinions and stakeholders in an explorative way. For this purpose, the
method of semantic network analysis seems very appropriate, as it combines quanti-
tative and qualitative methodologies. Although researchers select and quantify several
key words based on their theoretical foundation, it is only by their links to each
other that the meaning of the text becomes apparent (Segev, 2020). Therefore, the
method allows to confirm pre-existing ideas but also to explore and get new insights.
Based on our theoretical assumptions grounded in the network agenda-setting approach
and associative network theory, we are interested in what associations and topics arise
with respect to an (inter)national issue. Further, semantic network analysis is well
suited for the analysis of big data sets, as we are planning to do in this study.
We study international public opinion by looking at two high-reach newspapers
in three countries, namely Germany, France, and Ireland. The selection of these
countries is motivated by different reasons. First, these countries are among the top
five trading partners of the United Kingdom (Fleming, 2019) and are particularly
affected by the consequences of Brexit. Second, they differ from each other in
terms of their geopolitical proximity to Britain. France and Germany in their role
as EU founding countries are known as the driving forces of the European Union
and Ireland as the country with the highest geographic and cultural proximity to
the United Kingdom among the three. Indeed, Ireland may be directly affected: In
the last months, the Brexit referendum sparked, for instance, many discussions
around Northern Ireland’s borders.
 Network of cleavages? 119

The selected newspapers were chosen based on their reach and included the Irish
Independent and the Irish Times for Ireland, the Süddeutsche Zeitung and Bild Zeitung
for Germany, and Le Figaro and Le Monde for France. The news articles were
retrieved with the database Factiva. The sample included all articles published from
January 31 to February 1, 2020 (official UK exit date from the EU), which con-
tained the word Brexit in the title or the lead paragraph. This included 47 Irish
articles, 21 German articles, and 21 French articles. For the French sample, only
one of the two newspapers (Figaro) reported on the issue.
To study the social media discourse, we looked at posts that mentioned Brexit
on Twitter. Twitter is the most used micro-blogging website for public diplomacy
activities, which also explains why some authors speak of “Twiplomacy” when it
comes to digital diplomatic relations (Sotiriu, 2015). Further, Twitter is very well
suited for semantic network analysis and the study of user-generated content, as
posts are mainly made of texts in comparison to other social media like Instagram
or YouTube and users can interact without having to establish a following/fol-
lower/friend relationship.
The data were collected using the free extension for Google spreadsheets TAGS
v6.1, which connects to the Twitter API and enables the retrieval of large data sets
from Twitter with hashtags or specific Twitter profiles (Hawksey, 2019). As pro-
posed by Bjola et al. (2020: 409), Big Data analysis “may be used to examine the
words most commonly associated online with another country”. For this, hashtags
are a useful tool, as they tag and organize content on social media. By looking at the
co-occurrences of hashtags and words in the same tweet scholars can get insights on
public issues (Bjola et al., 2020; Chang, 2010; Sevin & Ingenhoff, 2018).
We decided to collect data around the hashtag #brexit, as this is actually the
main issue connected nowadays with the United Kingdom and challenging inter-
national relations. We analyzed tweets in the same timeframe as in the traditional
media analysis to make comparisons between the media channels possible. Further,
Twitter content is changing rapidly and is very event-driven (Andéhn et al., 2014).
Therefore, it is important to focus on a date that has a special significance, and we
hypothesize that this special point in time will bring up the different positions
about Brexit during the last months.
In total, 83,611 tweets were collected between January 31 and February 1. In
order to clean the data from retweets and bots, we deleted 55,696 duplicates. Out
of the remaining 27,913 tweets, we took a sample of 2000 tweets, which were
chosen randomly. We decided to analyze a smaller sample of tweets to make our
networks more readable and because previous studies have shown that smaller
Twitter samples are representative enough in highlighting the main topics discussed
(Chariatte et al., 2019).
Unlike newspaper articles, tweets were not examined separately by the country
of origin. The reason for this is that digital platforms cross national borders, and
opinions from a wide range of countries appear together. This is particularly true
for the hashtag “#brexit”, which is international and is used in many languages by
a wide range of users around the world.
120 J. Chariatte and D. Ingenhoff

After the sample was determined, its content was prepared for semantic network
analysis. The first step was to find out which words occur most frequently in the
text, while excluding irrelevant words and stop-words. Stop-words are recurring
words such as pronouns, adverbs, and prepositions, which differ for each language
and do not provide relevant meaning for the analysis. For the analysis of the media,
background information such as the author’s name and the newspaper section was
excluded, whereas on Twitter data, links, images, or profile information were not
considered.
In order to choose relevant words or actors in the frequency list, we looked
at their context and meaning in the original news articles and tweets. In total,
we selected 82 German words, 85 French words, and 100 Irish words. For the
Twitter data, we identified 92 relevant words. With these words, we built
networks of word pairs depending on their co-appearance in the same sentence
(for the news media) or in the same tweet (for Twitter, as tweets are often
short). The construction of the networks was done with Visone network ana-
lysis and visualization software (Brandes & Wagner, 2004). The analysis was
made in the respective language and then translated into English for the visua-
lization of the networks. Further, the size of the nodes was adapted according
to their degree centrality, whereas the number of links between pairs of words
(their co-occurrences in sentences or tweets) was shown by the thickness of
their links.

Brexit in the digital Twitter-sphere

In a first step, we examine the discourse on the British exit on Twitter. Twitter
discourse is characterized by the presence of various national and international
actors that interact with each other. As can be seen in Figure 6.1, the online dis-
course around Brexit is strongly characterized by populist ideas and is polarized,
with pro- and anti-European voices opposing each other. This is already clear from
the three central terms “people”, “EU”, and “UK”, but also from the seven clus-
ters calculated using the Louvain method.
Cluster 1 around the central term “people” is better and illustrates the motiva-
tions of the British referendum vote. As the word “working” indicates, this is a
decision largely supported by the working-class, which feels lost and abandoned in
the perceived dark times of globalization. Similar to the nationalist movements in
the US (“#maga”), they interpret Brexit as a positive return to independence, as in
the days of the British Empire (“pride”, “beautiful”, “happy”). Also, in the Cluster
2, we find positive evaluations: Brexit is associated with hope, a new acquisition of
freedom, power, and sovereignty, and interpreted as a sign of true democracy.
Further, we can observe a strong personalization with Nigel Farage or Boris
Johnson being cheered.
Brexit is also celebrated in Cluster 3, for instance, as a statement against the
European “leaders”, considered to be mostly French and German politicians
(“Ursula von der Leyen”, “Angela Merkel”, and “Emmanuel Macron”,
 Network of cleavages? 121

FIGURE 6.1 Semantic network of Tweets around #Brexit.

“Germany”, “France”). Calls for similar votes in Europe can be found across the
network (“#Frexit”, “#Italexit”, “#Nexit”, “#Polexit”)
At the same time, very critical comments on Brexit can be found in the clusters
presented. For example, the vote is assessed as a “shame”, a “blame”, a “disaster”, or
renamed as “Brexshit”. In Cluster 4, the topic of migration is addressed for the first
time by judging Brexit as a racist vote. Especially in Cluster 5, protests against the
Brexit vote are evident through the use of numerous hashtags (“#notmybrexit”,
“#stilleuropean”, and “#Iameuropean”. Particularly interesting is the fact that slogans
promoted by cities or regions are actually used in general Twitter usage. For exam-
ple, the Scottish Government’s statement “#leavealighton” but also Sadiq Khan’s
“#LondonIsOpen” are used by Twitter users, indicating a successful co-creation of
the place brand. The fact that London is in the middle of this pro-European bubble,
not only associated with business but also with values such as “respect”, suggests that
the capital has been able to position itself in the digital space as a political actor and
defender of values in its own right.
In line with the paradiplomacy thesis, another central sub-state actor is Scotland.
Cluster 6 reflects the discussion on the economic and political consequences for
nations of the United Kingdom but also for associated territories such as Gibraltar.
Also, the conflict arising around the Irish border is addressed. This is one of the few
clusters that is more informative than emotional.
The last cluster, Cluster 7, summarizes well the central ideas of the network.
The online discourse around Brexit shows a divided and polarized society
(“UK”–“divided”). While some celebrated Brexit as a victory for populism and
122 J. Chariatte and D. Ingenhoff

regaining control, other voices strongly condemned and mourned it (“mistake”–

“tears”–”sad”–“idiots”–“hate”) and showed solidarity with Europe. Information-
related terms, which can be found traditionally in the news are very few and limited
to some references to the economy (“bankers”), migration (“passport”), and new
relationship models (“Switzerland”). Interestingly, we can observe a certain distrust in
the media throughout the Twitter network (“#fakenews”, “lies”, “media”) and
dominance of emotional content.

Brexit in the international news

After having analyzed the social media discourse around Brexit on Twitter, we
explore whether the same voices and topics are to be heard and seen in interna-
tional news. Therefore, we chose to look at the international news of three
countries, namely Ireland, Germany, and France.
Figure 6.2 presents the semantic network of Irish news. Unlike the Twitter
network, it is less characterized by emotional comments but more focused on dis-
cussing related topics to Brexit. The word “Ireland” is central, suggesting that the
main concern here is the national impact of Brexit, and especially its implications
on Northern Ireland, which voted mainly against Brexit. In fact, the central words
“UK”, “trade”, “Northern”, and “border” point to the economic and political
challenges ahead. These are discussed in four different clusters. In Cluster 1, the

FIGURE 6.2 Semantic network of associations in the Irish media.

 Network of cleavages? 123

focus is on the financial consequences for Irish companies and workers in pro-
European Northern Ireland or in the border area. For example, farmers who pre-
viously received financial support from the EU are now demanding compensation
from the British government. Other issues include immigration and, above all, the
domestic security of the Irish Republic.
This is mainly taken up in Cluster 2: Around the central term “border”, fears are
raised from a hard frontier between Northern Ireland and Ireland, and the resur-
gence of riots that could endanger the Belfast Agreement. The word “border” is
further associated with the term “sea” and the central word “trade”. This points to
a potential border down the Irish Sea, which Dublin, or the Irish government, tries
to prevent by starting talks regarding future economic relations and tariffs.
This theme is further connected to Cluster 3, which deals with the negotiation
talks between Irish Finance Minister Donohoe and British Prime Minister Johnson
regarding new business deals and the numerous (northern) Irish jobs, which are put
at risk by the British referendum. These issues also influence the Irish elections,
which were taking place in Ireland at the time and are described in Cluster 4.
Campaigns of different political parties around the unity of Ireland, the “Border
Communities”, and the relations to the EU are addressed.
In total, we can observe that four of seven clusters deal only with consequences of
Brexit for (Northern) Ireland. The three remaining clusters focus on the international
sphere, the United Kingdom, and its different nations. In Cluster 5, the two central
terms “people” and “UK” can be identified. As in the Twitter network, the word
“people” may refer to the populist meaning of this vote. The terms associated with
“UK” show the dividedness of the nation and the fragility of the British identity:
While some promise Brexit as a sign of democracy, civil rights, and a sign of change,
others consider it as a crisis that would create damage and should be protested against.
Indeed, the commitment of politicians (McGuiness against Farage) but also of
cities to the European idea becomes apparent in Cluster 6. Thus London, which is
also the seat of the British government, is perceived as a world city that positions
itself as separate from England. Other cities and research centers such as Cambridge
similarly voted against the referendum. In Cluster 7, it becomes clear that Scotland
rejects the Brexit decision and is already making the case for an independence
referendum.
In general, we can observe that the whole Irish network depicts Brexit in a
negative way. When looking at the German and the French network, we can
observe a similar framing. Yet, unlike the Irish network, they are more concerned
about the consequences of Brexit for Europe and the United Kingdom than for
their own country.
In the German network (see Figure 6.3), we can observe seven clusters, and only
one specifically addresses the consequences for Germany. In Cluster 1, the
immense economic losses and difficulties of German (and especially Bavarian)
companies are addressed (“Germany”—“euro”—“billions”). However, it is inter-
esting to observe that the word “Germany” is nevertheless not the center of
attention, but rather the word “Great Britain”, which has many links to other
124 J. Chariatte and D. Ingenhoff

FIGURE 6.3 Semantic network of associations in the German media.

words (such as “French”, “England”, “British Currency”). This shows that more
than the national issue, German news is concerned by the international economic
consequences. In general, a lot of coverage is dedicated to economic and financial
topics, which is also evident from the fact that three other clusters deal with them.
For instance, Cluster 2 deals with monetary policy aspects, i.e., the handling of
central banks and national currencies, whereas Cluster 3 focuses on the stock
market and future economic relations. Interestingly, we can observe here that also
other rising issues such as the coronavirus come into play.
In Cluster 3, two central words are “London” and “Brussels”, which might be
explained by two reasons. First, cities play an important role in international relations
and serve as important trade centers (both London and Brussels are strongly connected
to “free trade agreements”). Second, capital city names are often used in the mass
media as synonyms for their respective national governments, which in times of Brexit
need to renegotiate their agreements during the insecure transition period.
Cluster 4 not only focuses on the economic but also political key issues Prime
Minister Johnson will have to handle. As the size of the nodes indicates, Johnson’s
main opponent will be the EU, with which he needs to discuss the exit deal and the
customs clearances in Dover. However, potential points of conflicts with Scotland,
the Irish government, or French Minister Barnier regarding fishing rights are also
visible. Finally, the connection between words like “contracts” and “USA” indicate
Johnson’s intention to reorientate and make new non-European trade agreements.
Cluster 5 is built around the central word “British”, which indicates that a strong
focus lies here on national identity. Brexit is interpreted as a way for Britons to
rediscover their national identity and regain control and sovereignty over their own
 Network of cleavages? 125

country, which they miss through supranational institutions like the EU. The
yearning for historical power like during the British Empire are voiced, and the
country’s distinctiveness as an island is also discussed. When looking at the news
articles, the term “island” is connotated more negatively, as it is pointed out that
today’s nations can no longer function as islands but are very dependent on other
states in times of globalization. Indeed, British public opinion is not exclusively
positive about Brexit; the population is divided. This division is considered as
“dark”, and there are concerns about hard borders, even talk of “trench warfare”.
This context also explains the relevance of the word “cohesion”, which is faltering
on the British side, while EU countries explicitly position themselves as one.
The position of the EU in the upcoming discussions becomes clear in Cluster 6.
German media covers how Europe will defend fundamental freedoms such as
freedom of movement and prevent the UK from cherry-picking. The word
“battle” indicates that the year of negotiations will be marked by tough talks.
Finally, pro-European actors in the UK appear in Cluster 7, who mourn the exit as
a mistake and try to correct it.
As said in the beginning, it can be observed that Brexit has numerous negative asso-
ciations (“mistake”, “fatal”, “farewell tear”, “damage”, “dark”, “getting drunk”,
“grieve”) in the German news. A similar picture can also be observed in the French news.
The French network (Figure 6.4) is made of six clusters according to Louvain and
is globally negative toward the British withdrawal referendum (“regret”, “shock”,

FIGURE 6.4 Semantic network of associations in the French media.

126 J. Chariatte and D. Ingenhoff

“disaster”, “pain”, “being wrong”). In Cluster 1, the position of Great Britain after
Brexit is critically examined. It is doubted that Great Britain will gain power through
Brexit, but rather will be weakened as a nation. For example, the British economy is
still dependent on European decision-makers, as the connection between the words
“economy” and “Von der Leyen” suggests. The United Kingdom is also isolated as a
research location, and there is talk of recruiting British researchers to France. The fate
of foreign citizens and workers in Britain is discussed.
Sub-national actors are quite present in the French network. First and foremost
is Scotland, which dominates the Cluster 2. The connection between the words
“Scottish”, “Sturgeon”, “stay”, and “nationality” point to the efforts of the Scottish
First Minister to make Scotland an independent nation. Also visible are the former
Prime Minister, Tony Blair, who wants to maintain links with Brussels. The
rejection of Brexit is evident through terms like “rejection”, “battle”, or “protest”.
Sub-national actors are also visible in the Cluster 3, which discusses new European
partnerships. “Paris” and “Berlin” refer to the respective national governments,
who want to get closer to trading partners that used to cooperate with the UK,
such as Denmark, Sweden, and the Netherlands. This shows that Europe wants to
maintain its dense trade network and may develop independently of the United
Kingdom. Still, paradiplomatic activities from the UK seem to pay off: links to
pro-European cities such as Edinburgh and London, which are clearly identified as
partners, will continue to be shown. The impact of city actors can be further
observed in Cluster 4. Here, French city mayors have their say, emphasizing their
friendship and solidarity with the British people sharing the idea of a united
Europe.
In fact, values play a big role in this cluster. France, for example, is associated
with the defense of European values (“values”, “continent”, “war”), and is por-
trayed as a country that stands up against populist and nationalist decisions
(“people”, “regret”, “decision”, “Thatcher”). Cluster 5 also shows that Brexit is
not just a termination agreement, but also a symbol of the crisis of trust and
democracy.
Topics such as populism are also found in the last Cluster 6, as we can find
words like “populism” and “victory”. Further in the cluster, we can observe a
personalization of the Brexit discourse, with associations to Prime Minister Johnson
and Donald Trump. As an antithesis to this, one finds the term “European”, which
is connected to the words “Germany” and “agreement”. This relates to the
upcoming economic negotiation talks and that Germany’s influence will be
decisive.
In summary, compared to the Twitter discourse, the media coverage in all the
countries studied has fewer emotional and positive associations with Brexit. Fur-
thermore, as the first entries in the summary Table 6.1 also show, the countries
have different priorities in their reporting: While the immediate UK neighbor,
Ireland, reports mainly on the direct political and economic consequences for its
own country, France and Germany discuss the consequences for European and
British identity and for the EU’s political and economic relations.
 Network of cleavages? 127

TABLE 6.1 Top 20 words with the highest degree centrality in all four analyzed units

Twitter Degree Ireland Degree Germany Degree France Degree

(%) (%) (%) (%)
EU 6.63 Ireland 8.79 Great 7.96 UK 8.59
Britain
people 4.71 UK 8.1 British 7.01 Johnson 7.59
UK 4.62 trade 5.47 EU 6.21 Eur- 5.76
opean
#brex- 3.17 EU 3.63 Johnson 5.73 London 3.47
itcelebration
friends 2.69 border 3.4 Europe 3.98 France 2.83
London 2.69 people 3.12 negotia- 3.18 Brussels 2.74
tions
freedom 2.6 northern 2.94 England 3.03 Scottish 2.29
sad 2.4 minister 2.17 London 3.03 partners 2.29
Irish 2.12 negotia- 1.89 economy 2.87 people 2.19
tions
free 2.02 business 1.66 free trade 2.55 thatcher 2.19
congratula- 2.02 govern- 1.56 Brussels 2.55 Paris 2.19
tions ment
Scotland 2.02 Johnson 1.46 bank 2.39 Denmark 2.1
hope 2.02 relation- 1.35 euro 2.23 negotia- 2.01
ship tions
happy 2.02 work 1.33 billions 2.07 Nether- 2.01
lands
@bor- 2.02 sea 1.3 cohesion 1.75 agree- 1.92
isjohnson ment
#leavea- 1.83 eco- 1.25 cherry- 1.59 north 1.92
lighton nomic picking
@nigel_fa- 1.83 Europe 0.92 prevent 1.59 market 1.83
rage
lies 1.63 talks 0.87 firms 1.43 Sweden 1.74
power 1.63 difficult 0.84 central 1.43 Blair 1.65
bank
free trade 1.44 leave 0.82 damage 1.43 free trade 1.65

Discussion and implications

In the age of digital platforms and the growing production of data, it is important
for diplomacy officials to keep an overview of the various (paradiplomatic) actors
and issues discussed in different communication channels. Semantic network ana-
lysis is a powerful method to map the discourse and identify patterns and themes
related to Brexit in traditional and digital media from large datasets and compare
128 J. Chariatte and D. Ingenhoff

them with previous research. This kind of analysis allows us to draw conclusions
and develop new communication strategies.
First, the present research clearly shows how important it is to consider the needs
of the population in international relations, which is well illustrated by the example
of Twitter. In line with Flew (2020), it is evident that citizens feel lost in times of
globalization and want to regain control, and therefore seek to emphasize their
national identity. The British referendum and its social media presence is only one
example of a more general trend: Hashtags such as #Frexit or #Italexit support a
general Eurosceptic discourse as described by FitzGibbon et al. (2017). Consistent
with the findings of Engesser et al. (2017), our results support the notion that
digital spaces are well suited for the spread of populist discourses. As Dolea et al.
(2020) highlight, the presence of populism in the public sphere does not have to be
ignored, as it might have a great impact on the country image, but also on the
national identity. Cooper (2019) even goes a step further, suggesting traditional
public diplomacy has failed to look at domestic concerns and needs, which play an
important role in the context of international relations.
Interestingly, we found that social media discourse is much less about informa-
tion or facts, but more about feelings, personalization, and values. There is also
increased distrust in traditional media, as indicated by the discourse around fake
news. But above all, our findings are in line with the results of Duncombe (2019),
who highlights the relevance of emotions and identity on social media. In order to
improve their international communication, public diplomacy officials need to
consider more the domestic needs and opinions.
Indeed, Brexit has challenged the idea of British identity. The cluster analysis
shows that the United Kingdom is not perceived as a unity, but as a divided country
that does not support Brexit unanimously. Through expressing their opinions on
Brexit, people at the same time constantly debate the country image across varied
platforms. As Zaharna (2020) highlights, it is essential to look more closely at the
domestic public.
In terms of the “New Public Diplomacy” and paradiplomatic activities, such as
city diplomacy, the international communication of countries is shaped by a variety
of actors, including cities. London is clearly associated with pro-European values
through the use of numerous hashtags, such as “#LondonIsOpen”. The rise of city
diplomacy has several implications. First, cities are indeed acting as so-called value
defenders, as described by Leffel (2018). Second, London’s branding strategy is
further adopted and used by Twitter users. This in turn confirms the potential of
social media for the co-creation of place brands (Compte-Pujol et al., 2018; Dinnie
& Sevin, 2020). Indeed, citizens use social media platforms to communicate their
place identity if they feel misrepresented by national authorities (Casais & Mon-
teiro, 2019; Insch & Walters, 2018).
This is an important finding: As Rickly-Boyd (2012) describes, sub-state actors
are perceived as particularly credible and must therefore be taken into account by
public diplomacy actors in their communication campaigns. Paradiplomatic actors
not only help to shape the digital arena, but also attract the attention of the news
 Network of cleavages? 129

media: In all three countries studied, both cities and regional sub-state actors are
present: Scotland’s pro-European stance and its slogan “#leavealighton” are dis-
cussed, and London is also described, albeit to a lesser extent, as a global and open
city, distinct from the UK. It is interesting that “open” is often mentioned in the
context of trade, as this was also the original meaning of the #LondonIsOpen cam-
paign. Still, we need to be aware that capital city names can also serve as a synonym
for their respective governments in the mass media. Therefore, one should also
consider the peculiarities of each medium.
By means of semantic network analysis, it becomes clear that the discourse around
Brexit differs greatly in the social media and mass media arenas. While social media
discourse was emotionally charged and allowed negative but also many positive
voices on the referendum to be heard, the news coverage described the background
and consequences of the referendum and framed it in all analyzed countries critically.
It should be noted here that, although the present study was conducted with a lim-
ited number of outlets, and their political orientation was not considered, the results
are coherent with previous findings. Shortly after the referendum, the European
Journalism Observatory also found that Brexit was covered negatively in the inter-
national media, regardless of political orientation (Lees, 2016). Further, our results
indicate a certain consistency in media coverage over time.
This negative framing proves interesting as it confirms but also extends the
findings of Bijsmans et al. (2018). The study shows that the international mass
media supports the opinion of pro-European decision makers about Brexit. Word
associations such as “cohesion” and “prevent cherry-picking” make it clear that EU
countries present themselves in the upcoming Brexit negotiations as a solid entity,
fighting steadfastly for its values and interests. This is intended to discourage other
EU countries with similar movements to follow the British example. But this
makes it more difficult for the UK to assert its interests. This is an example that
illustrates well the assumptions of role theory and Oppermann et al. (2020), that
one’s role and actions are strongly influenced by others. As can also be read from
the network, states can no longer function as islands in today’s network society
(Castells, 2011).
In addition to this, the semantic network analysis also revealed differences in
media coverage. For example, the centrality measures showed that while France and
especially Germany focused on the consequences of the British decision for Europe
(which emphasize their role as the founding fathers of the EU), Ireland focused
mainly on itself. This is in line with Wanta (2018), who pointed out that the rele-
vance of an issue depends strongly on its significance for the country and its policies.
But above all, it highlights the need to look at cultural differences and their proxi-
mity to the United Kingdom. With high geographic and cultural proximity, Ireland
pays careful attention to the political and economic consequences around the Irish
border. In France, consequences for democracy and European values are discussed,
while in Germany economic and financial topics dominate. The topic of migration,
which can, for instance, be found in the social media sphere, is almost absent in
international news. Also, by means of the network analysis, topics have emerged
130 J. Chariatte and D. Ingenhoff

throughout the different countries, such as fishing rights, which have turned out to
be important in the subsequent negotiations.
In the context of network agenda-setting theory, this different setting and association
of issues are of primary importance, since it is on this media basis that public opi-
nion about a country and its related issues is significantly formed (McCombs &
Shaw, 1972; Zhang & Meadows III, 2012). For public diplomacy officials, it can be
important for future international talks and communication campaigns to know
which issues a country places particular emphasis on.
To conclude, it can be drawn that semantic network analysis is well suited for
listening processes. As described by numerous scholars (Cull, 2019; Di Martino,
2019), it is absolutely necessary for public diplomacy actors to identify and under-
stand relevant publics, their interests and current problems at an early stage. Using
the Brexit example, the relevance and perception of sub-states and actors and their
political attitudes could be shown, highlighting the multi-faceted image of a
country. The relevance of contextual factors such as culture, national identity,
media type, or social movements such as populism were shown, which are essential
for understanding the public.

Limitations and future directions

For the purpose of this research, two high-reach newspapers per country were
selected to analyze international news coverage. The results of the present study
could be extended by looking at other characteristics, such as political orientation,
newspaper type, or quality standards when selecting newspapers. As mentioned
earlier, our results regarding negative framing have been in line with previous
findings of the European Journalism Observatory in 2016, working with a broader
sample. However, particularly relevant in the context of network analysis, it would
be interesting to know whether other issues, topics and frames around Brexit or
populism would become apparent with a different and bigger newspaper selection.
This study has highlighted differences in media coverage between countries. For
example, Ireland, which is culturally and geographically very close to the UK,
discussed Brexit differently than the two EU founding states, Germany and France.
Looking at other countries that differ by geographic or cultural proximity (e.g., also
EU vs. non-EU states) would be desirable. It would also be interesting to com-
plement the results with an examination of domestic media, as domestic public and
national identity have proven to be particularly relevant.
The Twitter-sphere was examined based on a common hashtag (#brexit), and
no primary distinction was made between the origin of tweets. The reason for this
was to reflect the broad and diverse social media discourse on the topic, including
users from different countries and languages. For a more in-depth study, it would
be interesting to use network analysis to find out what associations certain user
groups have with an issue like Brexit.
Since the discourse on Twitter is event-driven and the meaning and relevance of
certain hashtags and topics change quickly, it would be relevant to extend the
 Network of cleavages? 131

study period, especially for long-term issues such as Brexit. Thus, changes and
developments of an issue could be observed. As it is common for hashtags to be
linked to each other on Twitter, it would also be interesting to observe spillover
effects regarding their content.

Tips and lessons for the use of semantic network analysis

1. In order to ensure a certain diversity of voices and topics in your Twitter

network, you need to filter out content from automated tweets that distort
the relevance of keywords. Therefore, before constructing the network,
check your data for bots. You can recognize bots, for example, by the fact
that their content is often an incomprehensible string of hashtags or that
the same statements are conspicuously frequent.
2. When collecting data around a hashtag be aware that they may be different
versions of a hashtag (such as #brexit or #Brexit) that you might want to
include. Create a list of all the relevant hashtags related to the topic in question.
3. When working with data in different languages, as we did in our study,
you need to adapt your list of stop-words. Every language has its own
stop-words.
4. If you are working with different types of media (here, for instance: tradi-
tional and social media), do not forget to adapt the unit in which you
analyze the co-appearance of words. When transforming a linear text into
a two-dimensional visual map, you need to define for example if you are
looking at the sentences, tweet, or news item level.
5. When constructing semantic networks of traditional news content, you
may download the data from databases such as Factiva or Nexis Uni.
However, before making the network, delete information related to the
newspaper, publication time, or the author.

References
Acuto, M., Morissette, M., & Tsouros, A. (2017). City diplomacy: Towards more strategic
networking? Learning with WHO Healthy Cities. Global Policy, 8(1), 14–22.
Acuto, M., Kosovac, A., & Hartley, K. (2021). City Diplomacy: Another Generational Shift?
Diplomatica, 3(1), 137–146.
Aldecoa, F., & Keating, M. (Eds.) (1999). Paradiplomacy in Action: The Foreign Relations of
Subnational Governments (Vol. 4). London: Psychology Press.
Alvarez, M. (2020). The rise of paradiplomacy in international relations. E-International Relations.
https://www.e-ir.info/2020/03/17/the-rise-of-paradiplomacy-in-international-relations/.
Andéhn, M., Kazeminia, A., Lucarelli, A., & Sevin, E. (2014). User-generated place brand
equity on Twitter: The dynamics of brand associations in social media. Place Branding and
Public Diplomacy, 10(2), 132–144.
132 J. Chariatte and D. Ingenhoff

Andersen, J. R. (1983). The Architecture of Cognition. Cambridge, MA: Harvard University Press.
Aronczyk, M. (2013). Branding the nation: The global business of national identity. Oxford:
Oxford University Press.
Asdourian, B., & Ingenhoff, D. (2020). A Framework of City Diplomacy on Positive Outcomes
and Negative Emotional Engagement: How to Enhance the International Role of Cities and
City/Mayor Branding on Twitter? In S. Amiri & E. Sevin (Eds.) City Diplomacy (pp. 83–109).
New York: Springer.
Bäckstrand, K., Kuyper, J. W., Linnér, B.-O., & Lövbrand, E. (2017). Non-state actors in
global climate governance: from Copenhagen to Paris and beyond. Environmental Politics,
26(4), 561–579.
BBC (2016a). EU referendum results. https://www.bbc.co.uk/news/politics/eu_referendum/
results.
BBC (2016b). EU referendum. Most London boroughs vote to remain. https://www.bbc.
com/news/uk-politics-eu-referendum-36612916.
Biddle, B. J. (1986). Recent developments in role theory. Annual Review of Sociology, 12(1),
67–92.
Bijsmans, P., Galpin, C., & Leruth, B. (2018). “Brexit” in transnational perspective: an
analysis of newspapers in France, Germany and the Netherlands. Comparative European
Politics, 16(5), 825–842.
Bjola, C., Cassidy, J. A., & Manor, I. (2020). Digital Public Diplomacy: Business as usual or a
paradigm shift? In N. Snow & N. Cull (Eds.), Routledge Handbook of Public Diplomacy (2nd
ed.) (pp. 405–412). London: Routledge.
Bjola, C., & Zaiotti, R. (Eds.) (2020). Digital Diplomacy and International Organisations:
Autonomy, Legitimacy and Contestation. London: Routledge.
Blay Arráez, R., Antón-Carrillo, E., & López Font, L. (2019). Londres tras el Brexit: la
campaña de comunicación #LondonIsOpen en defensa de la identidad de la ciudad
[London after Brexit: the communication campaign #LondonIsOpen in defense of the
city’s identity.] Revista Latina de Comunicación Social, 74, 263–284.
Brack, N., & Startin, N. (2015). Introduction: Euroscepticism, from the margins to the
mainstream. International Political Science Review, 36(3), 239–249.
Brandes, U. &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In
M. Jünger & P. Mutzel (Eds.), Graph Drawing Software (pp. 321–340). New York:
Springer-Verlag.
Brewer, P. R., Graf, J., & Willnat, L. (2003). Priming or framing: Media influence on atti-
tudes toward foreign countries. Gazette (Leiden, Netherlands), 65(6), 493–508.
Casais, B., & Monteiro, P. (2019). Residents’ involvement in city brand co-creation and
their perceptions of city brand identity: a case study in Porto. Place Branding and Public
Diplomacy, 15(4), 229–237.
Castells, M. (2011). The Rise of the Network Society (Vol. 12). London: John Wiley & Sons.
Chang, H. C. (2010). A new perspective on Twitter hashtag use: Diffusion of innovation
theory. Proceedings of the American Society for Information Science and Technology, 47(1), 1–4.
Chariatte, J., Segev, E., & Ingenhoff, D. (2019, July 7–11). Tweeting #Country: A comparative
analysis of country images as a new instrument for developing international communication strategy
[Presentation]. Annual Conference of the International Association for Media and Commu-
nication Research (IAMCR), Madrid.
Christensen, C. (2013). @Sweden: Curating a nation on Twitter. Popular Communication, 11(1),
30–46.
Collins, S. D., DeWitt, J. R., & LeFebvre, R. K. (2019). Hashtag diplomacy: Twitter as a
tool for engaging in public diplomacy and promoting US foreign policy. Place Branding
and Public Diplomacy, 15(2), 78–96.
 Network of cleavages? 133

Compte-Pujol, M., de San Eugenio-Vela, J., & Frigola-Reig, J. (2018). Key elements in
defining Barcelona’s place values: the contribution of residents’ perceptions from an
internal place branding perspective. Place Branding and Public Diplomacy, 14(4), 245–259.
Cooper, A. F. (2019). Adapting public diplomacy to the populist challenge. The Hague
Journal of Diplomacy, 14(1–2),36–50.
Criekemans, D. (2010). Regional sub-state diplomacy from a comparative perspective:
Quebec, Scotland, Bavaria, Catalonia, Wallonia and Flanders. The Hague Journal of Diplo-
macy, 5(1–2), 37–64.
Cull, N. J. (2019). The tightrope to tomorrow: Reputational Security, collective vision and
the future of public diplomacy. The Hague Journal of Diplomacy, 14(1–2),21–35.
Di Martino, L. (2019). Conceptualising public diplomacy listening on social media. Place
Branding and Public Diplomacy, 15, 1–12.
Dinnie, K., & Sevin, E. (2020). The changing nature of nation branding: Implications for
public diplomacy. In N. Snow & N. Cull (Eds.), Routledge Handbook of Public Diplomacy
(2nd ed.) (pp. 137–144). London: Routledge.
Dolea, A. (2018). Public Diplomacy as Co-constructed Discourses of Engagement. In K. A.
Johnston & M. Taylor (Eds.), The Handbook of Communication Engagement (pp. 331–345).
London: John Wiley and Sons.
Dolea, A., Ingenhoff, D., & Beju, A. (2020). Country images and identities in times of
populism: Swiss media discourses on the “stop mass immigration” initiative. International
Communication Gazette, 83(4), 1–25.
Duncombe, C. (2019). Digital diplomacy: Emotion and identity in the public realm. The
Hague Journal of Diplomacy, 14(1–2), 102–116.
Duran, M. (2016). Paradiplomacy as a diplomatic broker: Between separating differences and
engaging commonalities. Brill Research Perspectives in Diplomacy and Foreign Policy, 1(3), 1–56.
Engesser, S., Ernst, N., Esser, F., & Büchel, F. (2017). Populism and social media: How
politicians spread a fragmented ideology. Information, Communication & Society, 20(8),
1109–1126.
FitzGibbon, J., Leruth, B., & Startin, N. (Eds.) (2017). Introduction: Transnational and pan-
European Euroscepticism. In Euroscepticism as a Transnational and Pan-European Phenom-
enon: The Emergence of a New Sphere of Opposition (pp. 1–13). London: Routledge.
Fleming, S. (2019). These are the UK’s biggest trading partners. World Economic Forum. http
s://www.weforum.org/agenda/2019/11/brexit-trade-uk-eu/.
Flew, T. (2020). Globalization, neo-globalization and post-globalization: The challenge of
populism and the return of the national. Global Media and Communication, 16(1), 19–39.
Golan, G. J., & Himelboim, I. (2016). Can World System Theory predict news flow on
Twitter? The case of government-sponsored broadcasting. Information, Communication &
Society, 19(8), 1150–1170.
Hanna, S., & Rowley, J. (2019). The projected destination brand personalities of European
capital cities and their positioning. Journal of Marketing Management, 35(11–12), 1135–1158.
Harnisch, S., Frank, C., & Maull, H. W. (Eds.) (2011). Role theory in international relations
(Vol. 90). New York: Taylor & Francis.
Hawksey, M. (2019). TAGS. Twitter archiving through Google Sheets. https://tags.ha
wksey.info.
Holmes, M. (2015). Digital diplomacy and international change management. In C. Bjola &
M. Holmes (Eds.), Digital Diplomacy. Theory and Practice (pp. 27–46). London: Routledge.
Huijgh, E. (2019). Public diplomacy’s domestic dimension in the European Union. In J.
Melissen (Ed.), Diplomatic Studies: Vol 15. Public Diplomacy at Home. Domestic Dimensions
(vol. 15) (pp. 52–82). Leiden: Brill Nijhoff.
134 J. Chariatte and D. Ingenhoff

Ingenhoff, D., & Buhmann, A. (2019). Public Diplomacy: Messung, Entstehung und Gestaltung
von Landesimages [Public Diplomacy: Measurement, Creation and Formation of Country Images].
Köln: Herbert von Halem Verlag.
Ingenhoff, D., Calamai, G., & Sevin, E. (2021). Key influencers in public diplomacy 2.0: A
country-based social network analysis. Social Media+ Society, 7(1), 1–12.
Insch, A., & Walters, T. (2018). Challenging assumptions about residents’ engagement with
place branding. Place Branding and Public Diplomacy, 14(3), 152–162.
Kampf, R., Manor, I., & Segev, E. (2015). Digital diplomacy 2.0? A cross-national com-
parison of public engagement in Facebook and Twitter. The Hague Journal of Diplomacy,
10(4), 331–362.
Kuznetsov, A. S. (2014). Theory and Practice of Paradiplomacy: Subnational Governments in
International Affairs. London: Routledge.
La Porte, T. (2013). City public diplomacy in the European Union. In M. K. D. Cross & J.
Melissen (Eds.), European Public Diplomacy. Soft Power at Work (pp. 85–111). New York:
Springer.
Lees, C. (2016). Will it kill us or make us stronger? How Europe’s media covered Brexit. http
s://blog.politics.ox.ac.uk/will-kill-us-make-us-stronger-europes-media-covered-brexit/.
Leffel, B. (2018). Animus of the underling: Theorizing city diplomacy in a world society.
The Hague Journal of Diplomacy, 13(4), 502–522.
Lievonen, M., & Luoma-aho, V. (2015). Ethical hateholders and negative engagement. A
challenge for organisational communication. In A. Catellani, A. Zerfass & R. Tench
(Eds.), Communication Ethics in a Connected World. Research in Public Relations and Organi-
sational Communication (pp. 285–303). Bern: Peter Lang.
Lippmann, W. (1946). Public Opinion (Vol. 1). New York: Transaction Publishers.
MacDonald, S. (2016). The impact of Brexit on the UK’s reputation, influence and soft
power. Cultural Trends, 25(4), 280–286.
Manor, I., & Segev, E. (2015). America’s selfie: How the US portrays itself on its social
media accounts. In C. Bjola & M. Holmes (Eds.) (2015), Digital Diplomacy (1st ed.) (pp.
103–122). London: Routledge.
McCombs, M. E., & Shaw, D. L. (1972). The agenda-setting function of mass media. Public
Opinion Quarterly, 36(2), 176–187.
Melissen, J. (2005). The New Public Diplomacy. New York: Springer.
Melissen, J. (2011). Beyond the New Public Diplomacy. The Hague: Netherlands Institute of
International Relations’ Clingendael’.
Mendes, A. (2013). International Broadcasting and Social Media. Washington, DC: Public
Diplomacy Council.
Michelmann, H. J., & Soldatos, P. (1990). Federalism and International Relations: The Role of
Subnational Units. Oxford: Oxford University Press.
O’Boyle, J. (2019). Twitter diplomacy between India and the United States: Agenda-
building analysis of tweets during presidential state visits. Global Media and Communication,
15(1), 121–134.
Oppermann, K., Beasley, R., & Kaarbo, J. (2020). British foreign policy after Brexit: Losing
Europe and finding a role. International Relations, 34(2), 133–156.
Rickly-Boyd, J. M. (2012). Authenticity & aura: A Benjaminian approach to tourism. Annals
of Tourism Research, 39(1), 269–289.
Schiavon. (2019). Comparative Paradiplomacy. London: Routledge.
Segev, E. (2018). Googling the world: Global and regional information flows in Google
trends. International Journal of Communication, 12, 2232–2250.
Segev, E. (2020). Textual network analysis: Detecting prevailing themes and biases in
international news and social media. Sociology Compass, 14(4), 1–14.
 Network of cleavages? 135

Sevin, E. (2016). Branding cities in the age of social media: a comparative assessment of local
government performance. In M. Z. Sobaci (Ed.), Social Media and Local Governments (pp.
301–320). New York: Springer.
Sevin, E., & Ingenhoff, D. (2018). Public diplomacy on social media: Analyzing networks
and content. International Journal of Communication, 12, 3663–3685.
Šimunjak, M., & Caliandro, A. (2020). Framing# Brexit on Twitter: The EU 27’s lesson in
message discipline? The British Journal of Politics and International Relations, 22(3), 439–459.
Smith, H. J. M. (2019). City networks and paradiplomacy as global public policy. In D.
Stone & K. Moloney (Eds.) The Oxford Handbook of Global Policy And Transnational
Administration (pp 129–147). Oxford: Oxford University Press.
Sotiriu, S. (2015). Digital diplomacy: Between promises and reality. In C. Bjola & M.
Holmes (Eds.), Digital Diplomacy (pp. 47–65). London: Routledge.
Strauß, N., Kruikemeier, S., van der Meulen, H., & van Noort, G. (2015). Digital diplo-
macy in GCC countries: Strategic communication of Western embassies on Twitter.
Government Information Quarterly, 32(4), 369–379.
Szpak, A., Gawłowski, R., Modrzyńska, J., & Modrzyński, P. (2020). Independent players or
shadow compatriots. How did British cities deal with the Brexit process? European Plan-
ning Studies, 28(2), 340–356.
Tavares, R. (2016). Paradiplomacy: Cities and States as Global Players. Oxford: Oxford Uni-
versity Press.
Van der Pluijm, R., & Melissen, J. (2007). City Diplomacy: The Expanding Role of Cities in
International Politics. The Hague: Netherlands Institute of International Relations’
Clingendael’.
Wanta, W. (2018). Media influence on the public’s perceptions of countries: agenda-setting and
international news. In D. Ingenhoff, C. White, A. Buhmann, & S. Kiousis (Eds.), Bridging
Disciplinary Perspectives of Country Image, Reputation, Brand, and Identity (pp. 230–241).
London: Routledge.
Wanta, W., Golan, G., & Lee, C. (2004). Agenda setting and international news: Media
influence on public perceptions of foreign nations. Journalism & Mass Communication
Quarterly, 81(2), 364–377.
Zaharna, R. (2020). Communication Logics of Global Public Diplomacy. In N. Snow & N. Cull
(eds.) Routledge Handbook of Public Diplomacy (2nd ed.) (pp. 96–111). London: Routledge.
Zhang, C., & MeadowsIII, C. W. (2012). International coverage, foreign policy, and
national image: Exploring the complexities of media coverage, public opinion, and pre-
sidential agenda. International Journal of Communication, 6, 76–95.
7
SEXUAL ASSAULTS BLINDSIDED BY
POLITICS ON TWITTER
Semantic Network analysis of symbolic #MeToo
cases in Japan and South Korea

Saki Mizoroki and Bumsoo Kim

Introduction
What happens when women accuse powerful men of sexual abuse: feminist soli-
darity or misogynistic victim-blaming? This chapter explores Twitter narratives
after two major sexual abuse allegations made public in Japan and South Korea
(hereafter Korea). While an emerging body of literature investigates feminist online
activism and expressions, they are relatively limited to Anglo-American cultural
context or single-country studies. This chapter compares the two cases in East Asia
which share striking similarities: (1) the two women were allegedly sexually abused
by their seniors who hold powerful positions; (2) the criminal justice system failed
them; and (3) while they were featured as a symbol of the #MeToo movement in
mainstream media, they faced fervent backlash for publicly speaking up in 2017
and 2018, respectively. Semantic network analysis of the reaction displayed on
Twitter to the two cases revealed quite a similar pattern—namely, the politicization
of the issue—and also a stark difference—the lack of feminist solidarity in Japan.
We begin with an overview of the two cases.

Summary of sexual harassment cases in Japan and South Korea

Ito-, journalist in Japan

“The act of rape killed me from within”. In May 2017, a freelance journalist Shiori
Ito- (hereafter Ito-) was sitting in front of TV cameras, determined. In the press
conference, due to the sexual nature of the allegation and in order to protect her
privacy, Ito- only revealed her age, 28, and her first name. However, she chose not
to hide her face, which was an extremely rare and brave move in Japan. She spoke
out that she was sexually assaulted in April 2015 by a man who was more than 20

DOI: 10.4324/9781003120100-8
 Sexual assaults and politics on Twitter 137

years older than her. The accused, Noriyuki Yamaguchi (hereafter Yamaguchi), is a
veteran journalist and then the Washington DC bureau chief for Tokyo Broad-
casting System (TBS), a major television media conglomerate in Tokyo. He has
been well known for his political connection with the administration at the time
and published biographies of Prime Minister Shinzo Abe (Ueda et al., 2017).
Ito- explained in detail what happened that night with Yamaguchi. According
to her, Yamaguchi asked her out, then 26 years old aspiring journalist, to a dinner
in Tokyo to discuss a job opportunity. After having a couple of drinks, she lost
consciousness, only to wake up in a hotel room with a sharp pain the next
morning around 5 am. He was on top of her (Ueda et al., 2017).
Ito- turned to the criminal justice system to accuse Yamaguchi. Later that month,
she filed a criminal complaint with the police, and the police obtained a warrant of
arrest on the charge of incapacitated rape. Yet, Yamaguchi got away at the last
minute in June 2015 thanks to Itaru Nakamura, the head of the criminal investi-
gation division at the Tokyo Metropolitan Police Department, who intervened to
suspend the arrest (“Higai jose- ga kokuhatsu!”, 2017). In August 2015, the police
eventually sent papers to the Tokyo District Public Prosecutors Office on Yama-
guchi; However, the case was dropped on the grounds of insufficient evidence a
year later in July 2016 (Ueda et al., 2017).
Ito- did not give up. In May 2017, in an effort to reopen the case, she filed a
complaint with the Committee for the Inquest of Prosecution. She then held the
aforementioned first press conference, where she came public. In September 2017,
the committee concluded that the prosecutors’ decision not to charge Yamaguchi
was appropriate (“Bo-ko- higai uttaeru”, 2017). Doors closed to accuse him in the
criminal justice system.
Ito- persisted. To hold him accountable, a week later, she launched a civil lawsuit
with the Tokyo District Court against Yamaguchi, seeking 11 million yen in
compensation (Seya, 2019). In October 2017, the same month when the Harvey
Weinstein story broke, a major publishing company Bungeishunju- announced that
her book “Black Box” (Ito-, 2017), which recounts the violence, would release and
made her full name public for the first time (“Bo-ko- higai uttaeru”, 2017). She,
herself, then gave a second press conference. Her story attracted huge media
attention in many countries and was featured as the Japanese #MeToo movement.
Among all, the BBC aired her one-hour documentary “Japan’s Secret Shame” in
2018, which was partially translated into Japanese online (Jenkin & Woods, 2018;
“‘Nihon no himerareta haji’ Ito-”, 2018).
After providing occasional short counter comments, Yamaguchi made his posi-
tion clear publicly. Two days after her second press conference, he published a
long counterargument denying her allegations in a monthly magazine Hanada
(Yamaguchi, 2017). In February 2019, he went on to file a 130-million-yen
counter-suit against Ito- for defamation with the Tokyo District Court (The court
decided to consolidate the two cases) (Seya, 2019).
Ito- won the case. In December 2019, the court ordered Yamaguchi to pay 3.3
million yen in damages to Ito- and dismissed Yamaguchi’s counter appeal. “[She
138 S. Mizoroki and B. Kim

was] forced to have sex without contraception, while in a state of unconsciousness

and severe inebriation”, ruled the court (McCurry, 2019). He appealed to the
Tokyo High Court in January 2020 (“Moto TBS kisha”, 2020). In May 2019, he
filed a criminal complaint with the police on suspicion of slander and false com-
plaint against her, but the case was dropped in December 2020 (Seya, 2020).

Seo, prosecutor in Korea

Jihyeon Seo1 has been credited with being the first to spread the #MeToo move-
ment in Korean society. On January 29, 2018, a female prosecutor of the Seoul
district, Jihyeon Seo (hereafter Seo), appeared in a live interview on a mainstream
TV channel news show, JTBC News Room, and accused her senior colleague,
Taegeun Ahn (hereafter Ahn), of groping her. The testimony by a high-profile
incumbent official shook the nation and triggered the #MeToo movement in
Korea. Numerous women have since come forward to accuse powerful figures in
all walks of life of sexual violence, including a former presidential contender, a
world-renown film director, a prominent poet, and popular actors (Jung & Jung,
2018; Maresca, 2018).
According to her account, in October 2010, at the funeral of a fellow prosecu-
tor’s father, Seo and Ahn sat down next to each other at a dinner table. He got
drunk and started to stroke her waist and hips. She could not say anything to him,
she said. Fellow prosecutors at the same table witnessed the sexual harassment, but
none of them intervened to stop him (J. Jang, 2018).
Before appearing on TV, she sought justice within the agency. However, when
Seo consulted senior prosecutors, who promised to talk Ahn into apologizing, she
was reassigned to a lowly position in Tongyeong, a small town on the south coast.
Suspecting Ahn might pull strings behind the scenes, she filed formal complaints,
only to be reprimanded for dishonoring the agency (Park, 2018a).
Disappointed by the agency’s inertia and feeling powerless, she made her mind up
to go public. “When I decided to speak out on TV—which amounts to social sui-
cide—I was ready to resign and become a recluse for the rest of my life”, she told
AFP (Jung & Jung, 2018). Ahn denied the accusations, and he explained that as a
devout Christian he gave a confession to God in church and said that “I don’t
remember correctly what happened at the funeral, but I heard about the news from
the media. If it happened, I would apologize to her, prosecutor Seo” (Donga, 2018).
Her TV testimony also got the attention of the prosecution service. A special
investigative team was launched and charged Ahn with abusing his power to relo-
cate her to protect his reputation. While Ahn denied his charges, the Seoul district
court sentenced him to a two-year jail term in January 2019. The Seoul Central
District Court’s Appellate Division upheld the prison sentence. The courts have
acknowledged that the sexual abuse happened, but he was not charged with it
because the one-year statute of limitations had expired.
Yet, the legal battle did not end here. One month after Ahn was released on
bail, in January 2020, the top court ordered a retrial. In September 2020, the Seoul
 Sexual assaults and politics on Twitter 139

Central District Court overturned the original ruling and acquitted him of the
charges of abuse of power (Choon, 2020).

Literature review

Sexism and its theoretical ground

Sexism is a belief system that maintains the social order and domination over women
on the basis of gender. It is an ideology that still perpetuates in Japanese and Korean
societies in the 21-first century. In explaining the power of racism, Arendt (1973)
adequately defined that an ideology “claims to possess … the intimate knowledge of
the hidden universal laws which are supposed to rule nature and man” (p. 159). She
goes on to say that an ideology is a system “based upon a single opinion that proved
strong enough to attract and persuade a majority of people and broad enough to
lead them through the various experiences and situations of an average modern
life” (p. 159). This is true of sexism as well as racism.
Sexism is comprised of a set of sexist ideas, which appear in sexist practices. It is
sexist because it assumes gender binarism and roles. Sexist ideas are normative; they
are norms, assumptions, and expectations, accompanying an auxiliary verb
“should” or “should not” in everyday communication, and uphold the sexist belief
system. Those ideas are taught and learned through practices. Once they are
internalized, they become epistemic reality and lead people to believe they “feel”
those normative necessities. This internalization is convenient for the system to
keep running without disruption, i.e., ideas that challenge the system.
Drawing on a critical reading of the theoretical work on ideology (Althusser,
1971; Hall, 1985) and misogyny (Manne, 2017, 2020), sexism can be con-
ceptualized as the system, and sexist practices as its execution. They work in tandem;
sexism needs sexist practices to maintain the system, and sexist practices need
sexism to naturalize their actions. Sexism rationalizes, justifies, and validates sexist
practices and, in turn, those practices serve to nurture, maintain, reinforce, and
deepen the system. It takes two to tango and (re)establish the whole system.
Sexism encourages and incentivizes sexist practices, and it also discourages and
disincentivizes practices that violate and challenge sexist ideas, the two feedback
loops; to moral offenders and violators, they attack, punish, blame, silence, deny,
and neglect; to moral defenders and adherers, they reward, support, and valorize
(Manne, 2017, 2020). For instance, when sexual violence survivors come forward
and ask for justice, she tends to face sexist practices in full force. Her testimony
will be denied, and her outcry will be neglected; she will be blamed for her
action, including the very act of speaking up, in an effort to silence her (e.g.,
Ogawa, 2020). Furthermore, Banet-Weiser and Miltner (2016) suggest that the
fear of female encroachment into previously almost exclusively white male spaces
drives some men to commit “popular misogyny, at its core, a basic anti-female
violent expression that circulates to wide audiences on popular media platforms”
(p. 172).
140 S. Mizoroki and B. Kim

This ideological mechanism of sexism can explain rape culture (Herman, 1984), for
example: One sexist idea in the belief system is that “aggressive male sexuality is ero-
ticized” and perceived as healthy and therefore encouraged (Keller et al., 2018). This
idea can easily lead to a more troubling idea that sexual assault is a natural part of sexual
relationships and even desirable, which allows the norm adherers to get away with it.
Victim-blaming is the flip side of the same coin. The normalization of aggressive
male sexuality naturalizes victim-blaming and vice versa. Keller et al. (2018) discuss:

Women are … seen as deserving or provoking rape by failing to perform a chaste

femininity, or for sending out signals to men that they are “up for it” … These
so-called “signals” include staying out late at night, drinking alcohol, flirting
with men, wearing “provocative” clothing, or being sexually active.
(pp. 23–24)

Sexism and feminist activism on social media

This general grasp of sexism, together with its execution mechanism and exploita-
tive nature, gives us a solid grounding to analyze sexist discourse on social media.
Scholarly exploration regarding sexist discourse on social media can be categorized
into two major directions: (1) examining gender-based hatred and aggression
toward women; and (2) investigating feminist counterarguments against sexism,
mainly hashtag activism.

Gendertrolling
Scholars have named a barrage of hostility, harassment, abuse, and violence directed
toward women in cyberspace many terms such as “gendertrolling” (Mantilla, 2013),
“e-bile” (Jane, 2014a, 2014b), “technology-facilitated sexual violence and harass-
ment” (Henry & Powell, 2015), “networked misogyny” (Banet-Weiser & Miltner,
2016), and “gendered cyberhate” (Jane, 2020). New cyber communication plat-
forms—namely, blogs and social media—spurred a massive flow of user-generated
content with constant interaction. Unfortunately, they also provided sexist moral
defenders and adherers with opportunities to practice sexist ideas online with mini-
mal cost and even access to much wider audiences who appreciate and support their
sexist practices beyond their immediate social circles. The new technology, indeed,
increases the visibility and engagement of online aggressions and, thus, aids their
collective effort to achieve a systemic goal, i.e., sustaining sexism.
Gender-based aggression toward women, however, is nothing new in nature. It
is just an articulation of traditional sexism in society adapted to this new cyber
platform. Banet-Weiser and Miltner (2016) cogently argue that this online aggres-
sion is still a product of the widespread, entrenched, and naturalized misogyny in
the Western culture itself. The literature in this orientation tends to illustrate,
categorize, or theorize changing aspects of sexist practices enabled by connective
technology while referring to existing research on sexism in society in general.
 Sexual assaults and politics on Twitter 141

Hashtag feminism
The other growing body of research examines a range of social media tactics—
hashtag activism—which make visible, problematize, and challenge sexism (see, for
instance, Baer, 2016; Clark, 2016; Keller et al., 2018; J. Kim, 2017; Mendes et al.,
2018). Social media, especially Twitter, gave rise to an entirely new mode of digital
engagement in activism. Users can post their personal stories, experiences, and
opinions, anonymously if they prefer, from anywhere in the world with a minimal
communication cost. Those voices hold the potential to be heard beyond their
existing networks and propagate across the platform further as they acquire more
engagements—reactions, comments, and shares—among users. Through this
interactive amplifying process, individual voices can grow into a larger collective
chorus and become highly visible as collaborative events—often with a catchy and
relatable hashtag—which spurs more users to jump onto the bandwagon. Primarily,
engagement characterizes this digital activism with its massive amplifying power.
Digital activism against sexism is known as hashtag feminism or feminist hashtag
activism (e.g., Clark, 2016). This novel digital engagement is worth scholarly ana-
lysis in four critical ways. First, it can unearth women’s collective voice, which
potentially serves to dismantle sexism. Those voices have not been taken seriously
in the mainstream media, historically marginalized and oppressed under the sexist
ideology (Benedict, 1992; Meyers, 1997). The sheer visibility of oft-silenced voices
gives users the courage to speak up, share their own experience, and join this
solidarity action of calling out sexist practices, or “feminist meme event” (Thrift,
2014). As in the case of #BeenRapedNeverReported, #YesAllWomen, and
#MeToo, the torrents of mediated voices dominated Twitter and subsequently
were reported by traditional media to the wider general public. This collective
storytelling successfully made visible persistent sexist practices in everyday life.
Thus, it serves to expose dominant sexism and implement alternative discourse that
is equal and fair.
Second, a hashtag, the key to digital engagement, acts as a reference point to
connect personal testimonials. It serves as an “interpretive frame” (Bonilla & Rosa,
2015) or a narrative frame of reference which people use to highlight and con-
ceptualize the specific aspects of their deeply personal experiences and to share
them. In other words, those testimonials, situationally diverse yet structurally simi-
lar, are customized to fit the frame. “[E]asy-to-personalize action themes” (Bennett
& Segerberg, 2012) mediate the feminist meme events.
Third, digital engagement fosters “affective solidarity” (Hemmings, 2012) that
connects participants in the hashtag feminism. Hemmings contends that it rises
from dissonance; the dissonance emerges between what sexist ideas instruct you
(not) to be or do and what you may be or act outside the dominance of ideol-
ogy—alternate possibilities. She goes on to say that the dissonance evokes affects
such as frustration, disgust, resentment, and rage. Twitter is quite handy and suited
for documenting the personal dissonance-driven affects in everyday life. Indeed,
the personal dissonance-driven affects, due to their very nature, are relatable and
142 S. Mizoroki and B. Kim

sticky and thus easy to spark engagement, connect participants, and build solidarity.
Collectively, people become more cognizant of alternate possibilities, and more-
over they become aware of the systemic and structural nature of what they thought
personal struggle, triggering a sense of injustice in society. Twitter acts as a platform
that builds affective solidarity among hashtag participants.
Forth, digital engagement is effective in calling out sexist practices and provide
counter-narratives to rectify the hegemonic system (Mendes et al., 2018). Pre-
dicated on a clear understanding of the structural nature of oppression, participants
use Twitter to call out people when they engage in sexist practices, or brandish
their gendered norms, expectations, moral codes, and sense of social order. More-
over, Twitter provides participants with “sites for collectively constructing counter-
narratives” (Bonilla & Rosa, 2015). Once the escalation cycle kicks in—visibility,
conceptualization, dissonance, affects, and solidarity—a call for change and a
counter-narrative to dominant sexism keep snowballing on digital platforms. News
coverage on this movement by online and traditional media accelerates the cycle,
spreading awareness among the general public beyond social media.

Research questions
The abovementioned literature indeed gives a strong argument behind the rise of
the digital phenomenon. However, they tend to focus on Western society except
for a few (e.g., Hassan et al., 2019; J. Kim, 2017; J. Kim, 2018; O’Mochain, 2020)
even though Japan has developed into the second biggest market for Twitter after
the US (Twitter, 2021) and has 45 million monthly active users (MAU) (Twitter
Japan, 2017). Twitter, which 38.7% of the population uses, is more popular than
Facebook in Japan and indeed the second largest social media in Japan after LINE
(Fujii & Hinuma, 2019; LINE, 2021; Ministry of Internal Affairs and Commu-
nications Japan, 2020).2 The use of social media has reached 87% in South Korea,
and Facebook was most favored by younger generations (1.9 million) and Twitter
with 0.8 million. The average amount of time spent on Twitter reached 594
minutes each month, followed by Facebook with 567 minutes and Instagram with
441 minutes (Shim, 2020). Moreover, most research pertains to the #MeToo
movement in Japan or Korea focus on media portray of the movement (e.g.,
Hasunuma & Shin, 2019; Lee, 2019; Lee, 2020; Starkey et al., 2019) rather than
qualitative or quantitative analysis of narrative on Twitter. Thus, by comparing
Twitter discourses on sexual violence in two languages, this chapter contributes to
the existing academic work on online personal expressions and feminist collective
action. To guide our analysis, the following research questions were posed:

RQ1: What specific patterns emerged in the Twitter discourse after the public
exposure of sexual violence allegations that gained notoriety in light of the
#MeToo movement?
RQ2: How did the support for the survivors and collective women’s voices
develop?
 Sexual assaults and politics on Twitter 143

Methods

Sample
To understand how personal expressions were displayed on Twitter after the two
major sexual allegations were made public, we gathered tweets by using the names
of survivors and assailants: Shiori Ito- (“伊藤詩織”) and Noriyuki Yamaguchi (“山
口敬之”) for the Japanese tweets; Jihyeon Seo (“서지현”) and Taekeun Ahn
(“안태근”) for the Korean tweets.
The data collection period for this study was set as follows: Japan (May 29,
2017–January 18, 2020) and Korea (January 29, 2018–January 18, 2020).3 Based on
the literature that collects prior tweets retroactively (Jungherr, 2016; Norheim et
al., 2018), we used a Python Script labeled GetOldTweets (https://github.com/Jef
ferson-Henrique/GetOldTweets-python) that is operated via the Twitter API to
collect tweets, and then we cleaned four datasets by deleting duplicates. This sam-
pling process yielded four datasets of tweets: 145,830 tweets mentioning Shiori Ito-,
129,960 tweets mentioning Noriyuki Yamaguchi, 17,838 tweets mentioning
Jihyeon Seo, and 5147 tweets mentioning Taekeun Ahn.
Furthermore, we filtered the datasets by deleting all tweets that contain URLs in
order to examine what users tweet with their own words as clearly as possible.
Approximately 60% of the tweets in Japanese and 65% in Korean included links to
other websites, i.e., news outlets, non-profit organizations, and political advocacy
groups. Those tweets usually have the autogenerated titles of the online articles that
automatically appear when you click the share button on the news website, which
can skew the results. Since it was not feasible to inspect tweets manually, we used
the presence of a URL as a proxy and deleted them altogether. The number of
tweets without URLs is 60,100 for tweets mentioning Ito-, 50,707 for Yamaguchi,
6531 for Seo, and 1550 for Ahn. The cleaned network maps better reflect user-
generated content because they actually tweet on their own or at least proactively
retweet other user-generated tweets.
Hence, a network that includes all tweets may reflect the general frames of news
organizations, and not only the personal opinions and comments of users. Deleting
all tweets with URLs is an efficient way to further focus on users’ personal tweets
written in their own words.

Semantic network analysis

Next, we conducted a semantic network analysis based on the top 100 keywords
in each dataset and constructed a network of word pairs appearing in the same
tweet. For the Japanese datasets, the network analysis procedure was carried out as
follows. First, we used KH Coder (Higuchi, 2016, 2017) to tokenize the words
and identify the top 100 frequently appeared words in each data set (see Segev,
2020). Second, inspecting the word list manually, we selected combinations of
words (morphemes) to be forcibly extracted as a single term. That is to say, since
144 S. Mizoroki and B. Kim

one Chinese character sexual (“性”) is central to our analysis, we pulled a combi-
nation of sexual and its frequently associated words such as violence (“暴力”),
damage (“被害”), and crime (“犯罪”) as one term. The documentary featured Ito-,
“Japan’s Secret Shame” (“日本の秘められた恥”), and her book, Black Box, were
also extracted as one term as well as personal names. Second, from the top of the
frequency list, we deleted less meaningful words for analysis, such as negative
auxiliary verbs and POS that consist of only hiragana (Higuchi, 2016, 2017). Third,
we manually searched within the top 1000 words and added to the list the syno-
nyms of the top 100 keywords. Moreover, the same words that were scripted dif-
ferently with Chinese character kanji topography and katakana syllabary, and the
same words with and without a hashtag were also added to the list. Examples
include Abe [安倍–アベ] and [#metoo–metoo]. The resulting number of the
keywords in the Ito- and Yamaguchi data were 107 and 116, respectively. We cal-
culated the number of co-occurrence of those keywords in the same tweet using R
script (using the quanteda package, see also Benoit et al., 2018).
We applied a parallel process to extract the top-100 keywords for Korean data-
sets using the same R script (Benoit et al., 2018). The network analysis procedure
was carried out as follows. First, we excluded prepositions, conjunction, and pro-
nouns from the selected keywords (e.g., “나 [I], “그래서 [so]”, 또한 [also]”).
Second, we combined synonyms from the list (e.g., 고맙습니다 [thank you] and
고마워 [thanks]) and then, we finalized 100 keywords, and these keywords were
used to generate two network maps.
We employed the network software Visone4 to generate network maps. Since each
initial network map was dense and hard to interpret, we merged synonyms and different
spellings associated with the central 100 keywords based on prior methods that show a
map simplification (Segev, 2020). For the Japanese datasets, we included nodes (key-
words) with at least 1100 links or more from both the female and male datasets. For the
Korean datasets, we included nodes with at least 18 links or more from the male dataset,
and nodes with 14 links or more from the female dataset. We also deleted isolated nodes
that were disconnected from the keyword networks as well as the most central key-
words that were used as the search terms themselves: Shiori Ito- (伊藤詩織) and Nor-
iyuki Yamaguchi (山口敬之) from the Japanese two datasets, Jihyeon Seo (서지현)
from the Korean female dataset, and Taeguen Ahn (안태근) from the Korean male
dataset. Louvain modularity was employed to generate clusters in a network map.

Results

Network analysis of the Japanese female dataset

Figure 7.1 illustrates the network of the words in a corpus of Japanese tweets that
were collected using the search term “Shiori Ito-”. The most frequently mentioned
words that co-occur appear as nodes. We found five distinct clusters that are clas-
sified by the Louvain method: Political cluster, Victim-blaming cluster, Judicial
cluster, Press Conference cluster, and Mixed cluster.
 Sexual assaults and politics on Twitter 145

FIGURE 7.1 Network of co-occurring words in Japanese tweets using the search term
“Shiori Ito-”.
Note. The size of a node indicates its betweenness centrality. The width of the ties indi-
cates the number of sentences in the corpus that mentioned each pair of words. The
grouping of the nodes are based on a cluster analysis using Louvain modularity. Network
visualization and calculations were performed with Visone (Brandes & Wagner, 2004).

First, the central node rape (レイプ, 強姦),5 bottom-left, is connected surprisingly
not to the sexual assault itself but a prominent political figure, Prime Minister, and thus
here constructs the Political cluster. It encompasses Abe (安倍), Shinzo (晋三), and
administration (政権) with which Yamaguchi reportedly holds a personal relationship (関
係). They are also mentioned together with media-related terms such as media (メ
ディア), its derogatory term mass media (マスコミ), and its report (報道).
A smaller, less central cluster at the bottom contains a couple of small nodes. In
an interview with BBC in the documentary—“Japan’s Secret Shame”—Diet (国
会) Member (議員) Mio Sugita (杉田) blamed Ito- for drinking a lot in front of a
man and losing her memory, and her statement (発言) was highly criticized as
victim-blaming (Jenkin & Woods, 2018).
A third cluster on the upper left corner represents the Judiciary theme with the
words trial (裁判) and litigation (訴訟) process. This cluster is quite descriptive;
Yamaguchi was not arrest[ed] (逮捕) because Nakamura (中村), who was the director
of criminal (刑事) investigations of the Tokyo Metropolitan Police Department,
reportedly intervened. Prosecutor (検察) also drop[ped] charge (不起訴) on the criminal
146 S. Mizoroki and B. Kim

(刑事) case. Ito- moved on to file a civil (民事) case, and court ruling (判決) judged that
she w[on] the case (勝訴).
The fourth cluster on the upper right encompasses her press conference (会見).
Most nodes are, again, descriptive and related to what she stated in her two press
conferences; she alleged that she received damage (被害)—sexual violence (性暴力)—
and moved on to accuse (告発) to pursue justice through the legal system.
A few exceptions are the word second (セカンド) and the feminist hashtag
#MeToo. Victim-blaming is called Sekando reipu (coined from a transliteration of
“second rape”, meaning a verbal repeated attack on a survivor) in Japanese, and this
term demonstrates concern about problematic abusive comments made on her.
#MeToo indeed appears in this cluster, but it does not accompany any terms that
indicate feminist solidarity, personal testimony, or call out.
The last cluster, the Mixed cluster, on the right, contains diverse terms, but two
frames were found: (1) regionality that contrasts world (世界) and Japan (日本) or
Japanese (日本人) and (2) gender that contrasts man (男性, 男) and woman (女性,
女). BBC refers to the channel on which the documentary was aired and it suggests
people’s concern about how foreign media portray Japan or Japanese in this
context.
This cluster shows the burgeon of a feminist chorus. Two words—human rights
(人権) and voice (声)—insinuate that users are more prone to mention the survi-
vor’s rights and the significance of breaking silence. Yet, these two words did not
appear to be central.

Network analysis of the Japanese male dataset

The semantic network map (Figure 7.2) of tweets that we produced from “Nor-
iyuki Yamaguchi” reveals distinctive discourse patterns: It is all about politics. The
Political cluster we found in Ito-’s map expands, the Judicial cluster leans more
toward politics, and Prime minister Abe (安倍) creates a central node.
What is so striking about these clusters is that the focus of the discourse diverged
from the sexual assault itself to the relationship between Yamaguchi and political
figures in the administration and prosecutor’s office, or even to two political scan-
dals that drew national media attention but had nothing to do with the allegation.
First, the Political cluster in the bottom-left corner clearly indicates the highly
politicized nature of the narratives. While closely connected to the term rape (レイプ,
強姦), this cluster is inundated with names in politics (政治): administration (政権),
Prime Minister’s office (官邸), Prime Minister (首相, 総理), Shinzo (晋三) Abe, which is
similar to an equivalent cluster appearing in Ito-’s network map.
Yet, this cluster indicated that the focus of the attention shifted specifically
toward two national political scandals of Abe—Moritomo (森友) and Kake (加計)
school (学園) scandal[s] (疑惑)—which have repeatedly made headlines in national
news since 2017. The Moritomo political scandal was exposed by the Asahi
newspaper. The Ministry of Finance (MOF) sold a state-owned batch of land to
the Moritomo school, where Abe’s wife Akie (昭恵) served as an honorary
 Sexual assaults and politics on Twitter 147

FIGURE 7.2 Network of co-occurring words in Japanese tweets using the search term
“Noriyuki Yamaguchi”.

principal, at a price that was way cheaper than the market price (Yoshimura & Ījima,
2017). Facing this political scandal, the MOF falsely announced that all related official
documents were already discarded and secretly forged internal documents to cover it
up. After the Asahi newspaper again exposed this cover-up (“Moritomo bunsho”,
2018), the MOF later publicly apologized for the forgery, and admitted that they kept
the original records, including the one that mentions the name of the Prime Minister’s
wife (“Zaimusho-, kakikae”, 2018).
The other political scandal also involves a school; Kake school, whose president
is a long-term friend (友) of Abe, was granted permission in 2017 to open a uni-
versity veterinarian department for the first time in 52 years in Japan. While Abe
denied all the allegations, the media exposed several internal documents suggesting
Abe’s political involvement in the permission process (“Shin gakubu”, 2017).
While Yamaguchi was not involved in those scandals, it is likely that Yamaguchi’s
widely publicized close connection with the Abe administration invoked close
association with those national political scandals.
Second, the Judicial cluster on the right includes scandals Yamaguchi reportedly
got benefited from. We see here two names: Nakamura (中村)—then the director
(部長) of criminal investigations of the metropolitan police department (MPD) (警視
庁), which is a police (警察) organization in Tokyo—who intervened to stop his
arrest (逮捕) on the charge of incapacitated rape, and Suga—then head chief (長官) of
cabinet secretariat (官房)—who was Nakamura’s former boss and reportedly helped
148 S. Mizoroki and B. Kim

Yamaguchi to get a job after Yamaguchi quit his job at TBS (“‘Jungo-kan taihojo-’ ga
nigiri tsubusare”, 2019).
Third, the Prime Minister cluster, which was built around Abe (安倍) recording
the highest score in-betweenness, also indicated that the focus of the attention
shifted toward politics: Diet (国会), Diet Member (議員), and ruling Liberal Demo-
cratic Party (LDP) (自民党) whose president was Abe. Attached to this cluster are
weekly (週刊) magazine Shincho- (新潮), which kept publishing hard-hitting article
(記事)-s on the political connection of Yamaguchi.
In addition to those clusters, we find another cluster at the bottom: the press
conference. This cluster mostly overlaps with the same cluster in the previous Ito-
network map but lacks #MeToo.

Network analysis of the Korean female dataset

Figure 7.3 portrays the network of frequent words in Korean for tweets that
mentioned the female prosecutor. Unlike Japanese tweets, it shows that the Korean
discourse in Twitter tended to appreciate and support her brevity for speaking up
to reveal her sexual harassment case (the cluster at the bottom). Her interview
(인터뷰) was perceived as brave (용기), which creates one of the central nodes.
They appreciated her exposure (폭로) of the case (사건) by simply writing thank you
(고맙습니다). They also showed direct support for her with terms such as cheer you
(응원합니다) and cheer up (힘내세요). The emergence of feminist solidarity can
be seen through words such as together (함께) and #Metoo (미투).

FIGURE 7.3 Network of co-occurring words in tweets in Korean using search term
“Jihyeon Seo”.
 Sexual assaults and politics on Twitter 149

In addition, the left cluster reflects the theme that Seo is not alone in fighting
against the misogynistic culture of the prosecution service (검찰). One of the central
nodes is the name of another female prosecutor, Eunjeong Lim (임은정), who
published a Facebook post to accuse the institution of its male-centered culture in
July 2017 (I. Jang, 2018). Calls for fairness (공정) and reformation of the public prose-
cution (검찰개혁) emerged on the map.
Finally, the sexual harassment case was clearly politicized. Kyoil Choi (최교일) is
Ahn’s former boss who held a position of the prosecutor’s chief of the Ministry of
Justice in 2010 and became a politician of a right-leaning Liberty Korea Party
(자유한국당)—now renamed People Power Party (국민의 힘). After the TV
interview, numerous people talked about Choi because when he was working in
the legal office of the attorney general, he reprimanded the sexual harassment
investigator (Eunjeong Lim, 임은정) and tried to close the investigation as soon as
possible (J. Jang, 2018). Also, the investigator Lim officially mentioned: “that the
prosecution service is a train with the ruptured braking system, and that the system
has to be reformed with eliminating the abusing authority inside the prosecution
service” (Cha, 2018). This triggered people to talk about the director of the prosecu-
tion service (검찰총장) and its culture and system. In contrast, some of the politi-
cians who are in liberal parties applauded her and mentioned, “[h]er courage will
be a valuable asset that advances our society of women’s dignity and gender
equality” (Jeong, 2018). The current Korean president Jaein Moon (문재인) was
among them and officially announced that “we should take the #MeToo move-
ment seriously, and the judicial authorities should actively investigate the case in
order to respond to the courage of the victims” (S. Kim, 2018).
Still, similar to the Japanese network map, the Twitter narrative was drawn into
a political scandal: After she spoke up, another female prosecutor Mihyun Ahn
(안미현), who worked at Chunchun branch office (춘천지검) revealed that she was
investigating the Kangwon land[’s] (강원랜드) employment corruption but was
pressured by the director of the prosecution service (검찰총장) and other politicians to
slow down (Heo, 2018). The sexual harassment and the political scandal triggered
Koreans to talk about the reformation of prosecution service (검찰개혁), which was
shown in the top left of the network map.

Network analysis of the Korean male dataset

As shown in Figure 7.4, the network map, which is based on the Korean male
dataset, has one large central node the prosecution service (검찰), which is the most
frequently mentioned keyword, and is closely associated with remained in custody
(법정구속), Kyoil Choi (최교일), and director of prosecution service (검찰총장).
Remained in custody (법정구속) was frequently mentioned because in 2019 Ahn,
who kept his position at the Ministry of Justice at that time, was sentenced to two
years in prison at the second trial for sexual harassment of prosecutor Seo.
Attention was also directed toward the body part Ahn molested, e.g., ass
(엉덩이). In 2010, prosecutor Seo said, “I was assaulted by prosecutor Ahn, who
150 S. Mizoroki and B. Kim

FIGURE 7.4 Network of co-occurring words in tweets in Korean using search term
“Taeguen Ahn”.

had served as the Minister of Justice at a funeral home on October 30, 2010. I was
promised to receive an apology from him, but I have not”. Specifically, she
revealed that “Ahn wrapped my waist and stroked my ass” (dongA.com, 2018). He
denied the accusation and told the reporters that he had made a confession to God
in church (교회).
Moreover, criticism for Ahn and his colleague is more rampant. Seo’s dis-
advantages (불이익) were related to his abusing authority (직권남용) and his personal
reprisal (개인 보복). Criticism was also targeted at ruined journalists (Gi-re-gi,
기레기). Journalism researchers are concerned that the practice of reporting sexual
harassment focused merely on the exposure of the case itself rather than scrutiniz-
ing and verifying it (Roh, 2018). The reason why people referred to ruined jour-
nalists in Korea is that even though there are journalists who strive to keep the
media fair and reliable, their efforts are often obstructed by media company owners
or managers. In Korea, the political partisanship of media outlets often invites cri-
ticism and leads to untrustworthiness in general, which drives people to call jour-
nalists who do not provide trustworthy reports, ruined journalists (Lim, 2014).
The map illustrates the attention to Seo’s TV interview. She appeared in JTBC
Newsroom on January 29 in 2018, which was hosted by Sukhee Son (손석희), a
famous TV anchor at that time in Korea. She said that “I came out to the TV
news show because I wanted to tell other sexual harassment victims that it was
not your fault. It took me eight years to realize it” (Park, 2018b). Support for Seo
can be found in terms such as brave (용기) and I support you (응원합니다) in this
map as well.
 Sexual assaults and politics on Twitter 151

Finally, the three clusters on the top of the network reflect the highly politicized
nature of the discourse. It features one conservative politician Kyoil Choi (최교일)
from the right-leaning Liberty Korea Party (자유한국당). Several political scandals
involving other former prosecutors were underway at the time the sexual harass-
ment case was revealed in 2010. The trial of the former senior secretary for civil
affairs and former prosecutor, Byoungwoo Woo (우병우), and the Kangwon land[’s]
(강원랜드) recruitment corruption (채용비리) occurred in 2012 and 2013. This
corruption case was closely associated with a congressman and a former prosecutor,
Sungdong Kweon (권성동) (Jung, 2017).

Discussion
In this chapter, we looked at how Twitter users discussed the widely publicized
sexual abuse allegations of two women in Japan and South Korea. Using semantic
network analysis, we identified three unique patterns in the Japanese network maps.
First, the most significant finding was the highly politicized nature of the Twitter
discourse. While the comparison between the networks showed a difference in
quantity, the political scandal indeed took over the focus of this rape allegation.
Second, the flip side of the dominance of politics was the absence of visible
women’s collective voice and action against sexism. While problematization of
victim-blaming was found, in contrast to #YesAllWomen (Thrift, 2014), #WhyIS-
tayed (Clark, 2016), or #Metoo (Mendes et al., 2018), the Ito-’s call for change did
not lead to a feminist meme event nor evoke an outpouring of affective solidarity
from survivors and supporters, let alone deconstruct sexist culture and initiate social
change. No prominent counter-narrative to sexual violence was developed (RQ2).
#MeToo appeared in the Political cluster but was not developed into a reference
point to produce its own cluster. The lack of meaningful feminist discussion among
Japanese Twitter users proved the points made by Bennett and Segerberg (2012):
“Personal action frames do not spread automatically. People must show each other
how they can appropriate, shape, and share themes”. Beyond this book chapter,
further research could examine why affective solidarity did not materialize.
Third, the mass media coverage played a significant role in shaping the Twitter
discourse, either on television (BBC) or in a weekly magazine (Shincho-), even
though this influence was much clearer in the network before filtering out tweets
with URLs, where news and news commentary were cited to add opinions and
feelings. Those reactions were directed toward not only what was reported but also
how the media portrayed the case. Commentary over mass media overshadowed
the women’s testimonials.
What do the silence of women, lack of personal story sharing, and absence of
online collective action mean? Feminist scholars have stressed, “hashtag feminism’s
discursive power is vital for activism surrounding violence against women, given
the ways in which popular discourse enables a culture in which sexual violence is
accepted as part of the norm” (Clark, 2016). Our research shows a darker picture
of discursive reality in current Japanese society.
152 S. Mizoroki and B. Kim

Compared to the Japanese networks, the Korean networks revealed both similarities
and differences in the digital discursive landscape. A striking resemblance was its poli-
ticization of the issue. Irrelevant political scandals and controversial political figures
were mentioned together, partially shifting the agenda from gender-based violence to
political debate. Still, the vital difference was that the Korean discourse included an
appreciation of and solidarity for Seo’s bravery, which did not emerge in the Japanese
case. Chandra and Erlingsdóttir (2020) claim that the courage was “contagious” and
“one of the most encouraging features” of the movement. We need to closely inspect
each tweet, of course, to see if it only shows appreciation for her courage or that her
courage inspired others to speak up. However, we can safely say our analysis illustrates
the very flint needed to spark the movement in the Korean case.
Moreover, the absence of nodes was telling in both cases. First, #Metoo did not
function as a narrative frame of reference to conceptualize and share personal
stories among Twitter users. Chandra and Erlingsdóttir (2020) contend that the
#MeToo movement was propelled by “individual appropriations of the hashtag”
and “the power of stories and personal narrative”. Although the two brave women
have been credited with being the key figures to spread the #MeToo movement,
our analysis showed that the hashtag failed to serve as “an easily personalized
storytelling prompt” (Clark, 2016) that mediated the feminist meme event.
Second, while value-laden words such as “cover-up” and “abusing authority” were
identified, most terms were quite descriptive, and no terms signified affects such as
resentment and rage. We cannot know what affects were expressed, if any, without
closely examining each tweet, but at least we contend that people did not use patterned
language to show their affects against sexist practices. The two cases did not produce “a
forum where anger and sorrow could be expressed in survivors’ own words and be
heard” (Chandra and Erlingsdóttir, 2020), which the #MeToo movement has achieved.
Why did the cases in Korea and Japan not inspire the #MeToo movement to the same
magnitude as in the Anglophone sphere? Was the “decades of collective work against
sexual abuse by groups and individuals” (MacKinnon, 2020) on which the movement
can spring out not enough? Are the “thresholds” of social support that people require to
“rebel or say what they actually think” too high (Sunstein, 2020)? Answering this ques-
tion is beyond the scope of this chapter, but this chapter demonstrates the lacuna of cas-
cading voices from the oppressed in Japan and Korea, to say the least.

Limitation
First, our analysis did not reveal how gender trolling played out over these issues.
While this research only used the formal names of four people, it is well known
that online users tend to deploy derogatory denominations to attack and blame the
targets in order to avoid consequences (Hamada, 2020). Additional data collection,
including these abusive pseudonyms, can reveal how people engaged in cyberbul-
lying and threw hateful comments on the targets.
Second, the engagement metrics, such as the number of likes and retweets without
comment, were not considered in this research. If we weighted the engagement, we
 Sexual assaults and politics on Twitter 153

would have shown a different landscape of how Twitter users reacted to these cases
and also interacted with each other. Further research can address this matter.
Finally, we only focused on the Twitter feed. While Twitter is one of the largest
social media platforms both in Korea and Japan, other social media platforms, such
as Facebook, Instagram, and YouTube, news curation websites such as Yahoo!
JAPAN, and blogging platforms such as Ameba blog and Note, are also widely
used by people to express their own opinions. Further research can include those
platforms to conduct a more comprehensive discourse analysis on the issue.

Tips and lessons from this process

1. Semantic network analysis is a powerful tool to examine large texts. In this

chapter, we produced networks of words based on all tweets mentioning
four people over almost two years (between 1550 to 60,100 tweets).
2. It is essential to carefully plan which search queries to select when collect-
ing data. In this case, we chose the names of four people. If we chose dif-
ferent keywords such as #Metoo in Japanese and Korean, we would have
found a different discursive landscape. It is also important to think thor-
oughly about how to clean the data. We decided to delete the tweets with
URLs that often refer to official news sources. In this way we could focus on
a greater proportion of user-generated content of the tweets without
URLs. Still, we found that the networks that included URLs were not much
different in terms of the main clusters and topics of discussion.
3. It is crucial to choose wisely which words to leave in the network to inter-
pret the discourse. We ended up deleting the names of four people (Ito-,
Yamaguchi, Seo, and Ahn), which appeared at the absolute center, in order
to have a clearer and meaningful observation of the network structure.

Acknowledgments
We would like to express our deepest gratitude to Professor Limor Shifman at the
Hebrew University of Jerusalem, who devoted her expertise and insight leading the
DigitalValues research project, from which this book chapter originated. Many thanks
also to Professor Kaori Hayashi at the University of Tokyo, who helped to develop this
study and whose mentorship profoundly enriched the process of writing this chapter.

Notes
1 Translation of Korean names is based on English spellings that appear on major news
outlets (i.e., The Korea Times).
2 Started as messaging app, LINE now provides news and digital money among other ser-
vices and counted as social media by the ministry of internal affairs and communications.
154 S. Mizoroki and B. Kim

LINE is used by 86.9% of the population (Ministry of Internal Affairs and Communications
Japan, 2020).
3 The data collection started on the date when the two cases were made public: Ito- held
her first press conference on May 29, 2017. Seo appeared on TV news program on Jan-
uary 29, 2018, and gave testimony. Data collection was terminated after the civil case
ruling in Japan and the supreme court ruling in Korea.
4 Available at: https://visone.info/
5 Throughout this chapter, we italicize the names of the nodes in the network maps and
put in parenthesis the names in original language for clarity.

References
Althusser, L. (1971). Ideology and ideological state apparatuses (notes towards an investiga-
tion) (B. Brewster, Trans.). In Lenin and Philosophy and Other Essays. New York: Monthly
Review Press. (Original work published 1970)
Arendt, H. (1973). The Origins of Totalitarianism. Cambridge, MA: Houghton Mifflin Harcourt.
Baer, H. (2016). Redoing feminism: Digital activism, body politics, and neoliberalism. Fem-
inist Media Studies, 16(1), 17–34. doi:10.1080/14680777.2015.1093070.
Banet-Weiser, S., & Miltner, K. M. (2016). #MasculinitySoFragile: Culture, structure, and
networked misogyny. Feminist Media Studies, 16(1), 171–174. doi:10.1080/14680777.2016.
1120490.
Benedict, H. (1992). Virgin or Vamp: How the Press Covers Sex Crimes. Oxford: Oxford Uni-
versity Press.
Bennett, W. L., & Segerberg, A. (2012). The logic of connective action: Digital media and
the personalization of contentious politics. Information Communication and Society, 15(5),
739–768. doi:10.1080/1369118X.2012.670661.
Benoit, K., Watanabe, K., Wang, H., Nulty, P., Obeng, A., Müller, S., & Matsuo, A..
(2018) “quanteda: An R package for the quantitative analysis of textual data”. Journal of
Open Source Software, 3(30), 774. doi:10.21105/joss.00774.
Bo-ko- higai uttaeru 28-sai jose- ga shuki shuppan, Shiori-san, se- mo ko-hyo- [a 28-year-old
woman who pressed charge on violence publishes a memoir, Shiori also made her family
name public]. (2017, October 13). Sankei Shimbun. https://www.sankei.com/affairs/
news/171013/afr1710130036-n1.html.
Bonilla, Y., & Rosa, J. (2015). #Ferguson: Digital protest, hashtag ethnography, and the
racial politics of social media in the United States. American Ethnologist, 42(1), 4–17.
doi:10.1111/amet.12112.
Brandes, U., &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In M.
Jünger & P. Mutzel (Eds.), Graph Drawing Software (pp. 321–340). New York: Springer-Verlag.
Cha, B. (2018, February 6). Im-Eunjeong, Josadan Chulseog “Seongchuhaeng-Eun Sangha
gwonlyeog Munje” [Im, Eun-jeong attends the investigation team “Sexual harassment is a
matter of upside-down and power”]. Yonhap News. https://www.yna.co.kr/view/
MYH20180206021600038.
Chandra, G., & Erlingsdóttir, I. (2020). Introduction: Rebellion, revolution, reformation. In
The Routledge Handbook of the Politics of the# MeToo Movement (pp. 1–23). London: Routledge.
Choon, C., (2020). #MeToo battle continues in South Korea, 2 years after the first case.
The Strats Times. https://www.straitstimes.com/asia/east-asia/metoo-battle-continues-
in-s-korea-2-years-after-first-case.
Clark, R. (2016). “Hope in a hashtag”: the discursive activism of # WhyIStayed. Feminist
Media Studies, 16(5), 787–804. doi:10.1080/14680777.2016.1138235.
 Sexual assaults and politics on Twitter 155

dongA.com (2018). Seojihyeon Geomsaga ‘Eongdeong-I Manjin Gahaeja’lo Jimoghan

Antaegeun Geomsa, Nugu? [Prosecutor Ahn Taegeun, whom Prosecutor Jihyeon Seo
pointed out as the “perpetrator who touched the ass”, who?]. https://www.donga.com/
news/Society/article/all/20180130/88424598/2.
Fujii, R., & Hinuma, S. (2019, July 8). Feisubukku Japan Hasegawa daihyo- ga kataru “tainin
no shin’i” [Facebook Japan Rep. Hasegawa talks “truth of retirement”]. CNET Japan.
https://japan.cnet.com/article/35139021/.
Hall, S. (1985). Signification, representation, ideology: Althusser and the post-structur-
alist debates. Critical Studies in Media Communication, 2(2), 91–114. doi:10.1080/
15295038509360070.
Hamada, K. (2020, January 8). “Ito- Shiori-san intabyu-” mangaka Hasumi Toshiko-shi ra wo
teiso [“Shiori Ito- interview” Cartoonist Ms. Toshiko Hasumi and others were sued].
Business Insider Japan. https://www.businessinsider.jp/post-214220.
Hassan, N., Mandal, M. K., Bhuiyan, M., Moitra, A., & Ahmed, S. I. (2019, January).
Nonparticipation of Bangladeshi women in #MeToo movement. In Proceedings of the Tenth
International Conference on Information and Communication Technologies and Devel-
opment (pp. 1–5). New York: Association of Computing Machinery. doi:10.1145/
3287098.3287125.
Hasunuma, L., & Shin, K. Y. (2019). #MeToo in Japan and South Korea: #WeToo,
#WithYou. Journal of Women, Politics & Policy, 40(1), 97–111. doi:10.1080/
1554477X.2019.1563416.
Hemmings, C. (2012). Affective solidarity: Feminist reflexivity and political transformation.
Feminist Theory, 13(2), 147–161. doi:10.1177%2F1464700112442643.
Herman, D. (1984). The rape culture. In Women: A Feminist Perspective (pp. 45–53). Cali-
fornia City: Mayfield.
Henry, N., & Powell, A. (2015). Embodied harms: Gender, shame, and technology-
facilitated sexual violence. Violence Against Women, 21(6), 758–779. doi:10.1177%
2F1077801215576581.
Heo, Y. (2018, February 4). [Dandok] “Gang-Wonlaendeu Chaeyongbili Susa Oeab”
Hyeonjig Geomsa Poglo [[Exclusive] Revelation of the incumbent prosecutor of
“Kangwon Land investigation of employment corruption, external pressure”]. IMBC.
https://imnews.imbc.com/replay/2018/nwdesk/article/4518583_30181.html.
Higai jose- ga kokuhatsu! “keishicho- ke-jibucho-” ga nigiri tsubushita “Abe so-ri” bettari kisha
no “Jungo-kan taihojo-” [A female victim accuses! “Arrest warrant on the charge of inca-
pacitated rape” for a pet reporter of “Prime Minister Abe” which was covered up by “the
head of the criminal investigation division at the Tokyo Metropolitan Police Depart-
ment”]. (2017, May 18). Shu-kan Shincho-, 62(19), 22–25.
Higuchi, K., (2016). A Two-Step Approach to Quantitative Content Analysis: KH Coder
Tutorial Using Anne of Green Gables (Part I). Ritsumeikan Social Science Review, 52(3), 77–91.
Higuchi, K., (2017). A two-step approach to quantitative content analysis: KH coder tutorial
using Anne of Green Gables (Part II). Ritsumeikan Social Science Review, 53(1), 137–147.
Ito-, S. (2017). Black Box. Chiyoda: Bungeishunju-.
Jane, E. A. (2014a). “Your a ugly, whorish, slut” understanding E-bile. Feminist Media Stu-
dies, 14(4), 531–546. doi:10.1080/14680777.2012.741073.
Jane, E. A. (2014b). Back to the kitchen, cunt: Speaking the unspeakable about online
misogyny. Continuum, 28(4), 558–570. doi:10.1080/10304312.2014.924479.
Jane, E. A. (2020). Online abuse and harassment. The International Encyclopedia of Gender,
Media, and Communication, 1–16. doi:10.1002/9781119429128.iegmc080.
Jang, I. (2018, December 28). Seojihyeon-I Ssoa Ollin Jag-Eun Gong [A small ball raised by
Jihyun Seo]. Sisa-In. https://www.sisain.co.kr/news/articleView.html?idxno=33508.
156 S. Mizoroki and B. Kim

Jang, J. (2018, January 30). Lim-Eunjeong Geomsa Peiseubug Gong-Gae Geul “Seojihyeon
Geomsa Seongchuhaeng Sageon” Naebu Gamchal Gwajeong Sangse Gong-Gae [Prose-
cutor Lim Eun-jung’s Facebook release, opening the case details of “Prosecutor Seo Ji-
hyun’s sexual harassment case”]. Segaeilbo. https://www.sedaily.com/NewsVIew/
1RUMMYCMCT.
Jenkin, E. (Producer & Director), & Woods, B. (Executive Producer). (2018). Japan’s Secret
Shame [Video]. London: True Vision.
Jeong, J. (2018, January 31). Simsangjeong Seojihyeonssi Inteobyuleul Bomyeo Jangjayeon
Sageon-Eul Tteoollyeossda [Shim, Sang-jung “I remembered the Jang, Ja-yeon case while
watching the interview with Seo, Jihyeon”]. Kyunghyang Sinmun. http://news.khan.co.
kr/kh_news/khan_art_view.html?art_id=201801311437001.
Jung, H., & Jung Y. (2018, September 28). SKorea’s quiet #Metoo trailblazer embraces
though fight. AFP. https://www.afp.com/en/skoreas-quiet-metoo-trailblazer-embraces-
tough-fight.
Jung, M. (2017, September 12). Kangwon Land admits massive corruption. The Korea Times.
https://www.koreatimes.co.kr/www/nation/2017/09/251_236357.html.
Jungherr, A. (2016). Twitter use in election campaigns: A systematic literature review. Journal
of Information Technology & Politics, 13(1), 72–91. doi:10.1080/19331681.2015.1132401.
“Jungo-kan taihojo-” ga nigiri tsubusare 4 nen… “Ito- Shiori-san” vs. “Abe kantei bettari
kisha” no ho-tei taiketsu [Four years after the cover-up of a warrant of arrest on the charge
of incapacitated rape … Legal battle of “Ms. Shiori Ito-” vs. “Abe administration’s pet
reporter”]. (2019, July 18). Shu-kan Shincho-, 64(27), 30–33.
Keller, J., Mendes, K., & Ringrose, J. (2018). Speaking “unspeakable things”: documenting
digital feminist responses to rape culture. Journal of Gender Studies, 27(1), 22–36.
doi:10.1080/09589236.2016.1211511.
Kim, J. (2017). #iamafeminist as the “mother tag”: feminist identification and activism
against misogyny on Twitter in South Korea. Feminist Media Studies, 17(5), 804–820.
doi:10.1080/14680777.2017.1283343.
Kim, J. (2018). After the disclosures: a year of #sexual violence in the film industry in South
Korea. Feminist Media Studies, 18(3), 505–508. doi:10.1080/14680777.2018.1456168.
Kim, S. (2018, February 26). Seojihyeon Geomsa Eung-Wonhaessdeon Mundaetonglyeong
“Mituundong Jeoggeug Jiji” [President Moon, who cheered for prosecutor Seo, Ji hyeon,
“Strongly support MeToo movement”]. E-daily. https://www.edaily.co.kr/news/read?
newsId=03644086619114456&mediaCodeNo=0.
Lee, J. (2020). Newspaper coverage of sexual assaults in South Korea: Before and after #MeToo
Movement. In Proceeding of the 3rd International Conference on Future of Women (Vol.
3, No. 1, pp. 1–11). doi:10.17501/26028646.2020.3101.
Lee, M. (2019). #MeToo and broadcast journalism in South Korea: The gatekeeping pro-
cess of #MeToo. Interactions. Studies in Communication & Culture, 10(3), 223–240.
doi:10.1386/iscc.10.3.223_1.
Lim, Y. (2014, December 24). Min-Eonlyeon Changlib 30junyeon Teugjib - Eonlonsiltaebogo
“Gilegi” [Min Eon-ryeon’s 30th Anniversary special report on media status “Gi-re-gi”].
Newstapa. https://newstapa.org/article/PJBZd.
LINE. (2021). LINE business guide (summary). https://www.linebiz.com/sites/default/
files/media/jp/download/LINE%20Business%20Guide_202107-12.pdf.
MacKinnon, C. A. (2020). Global #MeToo. In The Routledge Handbook of the Politics of the#
MeToo Movement (pp. 42–54). London: Routledge.
Manne, K. (2017). Down Girl: The Logic of Misogyny. Oxford: Oxford University Press.
Manne, K. (2020). Entitled: How Male Privilege Hurts Women. New York: Crown.
 Sexual assaults and politics on Twitter 157

Mantilla, K. (2013). Gendertrolling: Misogyny adapts to new media. Feminist Studies, 39(2),
563–570.
Maresca, T. (2018). South Kore’s MeToo movement topples high-profile politicians and
entertainment icons. https://www.usatoday.com/story/news/world/2018/04/03/south-
koreas-me-too-movement-topples-politicians-entertainment-icons/476895002/.
McCurry, J. (2019, December 18). Shiori Ito, symbol of Japan’s MeToo movement, wins
rape lawsuit damages. Guardian. https://www.theguardian.com/world/2019/dec/18/
shiori-ito-symbol-of-japans-metoo-movement-wins-lawsuit-damages.
Mendes, K., Ringrose, J., & Keller, J. (2018). #MeToo and the promise and pitfalls of
challenging rape culture through digital feminist activism. European Journal of Women’s
Studies, 25(2), 236–246. doi:10.1177%2F1350506818765318.
Meyers, M. (1997). News Coverage of Violence Against Women: Engendering Blame. SAGE
Publications.
Ministry of Internal Affairs and Communications Japan. (2020). Reiwa gan nendo jo-ho-t-
su-shin media no riyo-jikan to jo-ho-ko-do- ni kansuru cho-sa ho-kokusho [Fiscal year Reiwa 1
report on utilization time and information behavior on information communication
media]. https://www.soumu.go.jp/main_content/000708015.pdf.
Moritomo bunsho kakikae no utagai [Moritomo documents were possibly forged]. (2018,
March 2). Asahi Shimbun, 1.
Moto TBS kisha no Yamaguchi-shi ga ko-so [Former TBS reporter Mr. Yamaguchi appealed
in court]. (2020, January 7). Asahi Shimbun, 29.
“Nihon no himerareta haji” Ito- Shiori-shi no dokyumentarı- wo BBC ga ho-so- [BBC aired
Shiori Ito-’s documentary “Japan’s Secret Shame”]. (2018, June 29). BBC. https://www.
bbc.com/japanese/features-and-analysis-44638987.
Norheim, G., Boczek, K., Koppers, L., & Erdmann, E. (2018). Reuniting a divided public?
Tracing the TTIP debate on Twitter and in traditional Media. International Journal of
Communication, 12, 548–569.
Ogawa, T. (2020, August 20). Ito- Shiori-san, Sugita Mio-giin ra ni aratana sosho- teiki
ronten wa? [Shiori Ito- filed a new lawsuit against Diet member Mio Sugita among others.
What is contested?]. Yahoo! JAPAN News. https://news.yahoo.co.jp/byline/ogawatama
ka/20200820-00194226/.
O’Mochain, R. (2020). The #MeToo movement in Japan: Tentative steps towards transfor-
mation. In The Routledge Handbook of the Politics of the #MeToo Movement (pp. 360–371).
London: Routledge.
Park, S. (2018a, January 30). “Kkog Hago Sip-Eun Mal Issda” Seojihyeon Geomsaga JTBC
Naon Iyu [“I have something I want to say” The reason Prosecutor Jihyeon Seo came
out of JTBC]. Kookmin Ilbo. http://news.kmib.co.kr/article/view.asp?arcid=0012089374.
Park, S. (2018b, January 31). Seojihyeon “Ae Eommalo 8nyeongan Suchisim” … Seong-
chuhaeng Pogpung Hwimallin Geomchal [Seo, Jihyeon “I shame myself for 8 years as a
mother”…Prosecutors caught in a rape storm]. JoongAngIlbo. https://news.joins.com/a
rticle/22331997.
Roh, J. (2018, July 22). KBS “gilegi toechi” peulojegteu, ibeon-en “mitu bodo” bipyeong [KBS
“Eradication of Giregi (ruined journalists)” Project, This Time Criticized Me Too Report].
Media Today. http://www.mediatoday.co.kr/news/articleView.html?idxno=143734.
Segev, E. (2020). Textual network analysis: Detecting prevailing themes and biases in
international news and social media. Sociology Compass. doi:10.1111/soc4.12779.
Seya, K. (2019, December 18). Sho-so shita Ito- Shiori-san ga, hanketsu chokugo ni katatta
kotoba “keshiki ga izen toha mattaku chigau” [“It all looks different now” Shiori Ito- who
won the case said right after the ruling]. BuzzFeed Japan. https://www.buzzfeed.com/jp/
kensukeseya/shiori-ito-win.
158 S. Mizoroki and B. Kim

Seya, K. (2020, December 25). “Kyogi” to uttae rare teita Ito- Shiori-san, fukiso. Seihigai wo
meguri, sono omoi wa [Shiori Ito- who was accused of “a false complaint” was not
indicted. What does she think about sexual abuse?]. BuzzFeed Japan. https://www.buzz
feed.com/jp/kensukeseya/shiori-ito-18.
Shim, W. (2020, September 7). Korea’s social media penetration rate ranks third in world.
The Korea Herald. http://www.koreaherald.com/view.php?ud=20200907000815#:~:
text=South%20Korea‘s%20social%20media%20penetration,to%20a%20local%20report%
20Monday.&text=The%20rate%20of%20China%20was,percent%20and%2065%20percent%
2C%20respectively.
Shin gakubu “So-ri no iko-” [“Prime minister’s will” new department]. (2017, May 17). Asahi
Shimbun, 1.
Starkey, J. C., Koerber, A., Sternadori, M., & Pitchford, B. (2019). #MeToo Goes Global:
Media Framing of Silence Breakers in Four National Settings. Journal of Communication
Inquiry, 43(4), 437–461. doi:10.1177%2F0196859919865254.
Sunstein, C. R. (2020). #MeToo as a revolutionary cascade. In The Routledge Handbook of the
Politics of the #MeToo Movement (pp. 34–41). London: Routledge.
Thrift, S. C. (2014). #Yesallwomen as feminist meme event. Feminist Media Studies, 14(6),
1090–1092. doi:10.1080/14680777.2014.975421.
Twitter (2021, July 22). Q2 2021 letter to shareholders. https://s22.q4cdn.com/826641620/
files/doc_financials/2021/q2/Q2'21-Shareholder-Letter.pdf.
Twitter Japan [@TwitterJP]. (2017, October 27). Itsumo, soshite nan’nen mono aida,
Twitter wo goriyo- itadaki arigato- gozaimasu. Okage samade nihon deno gekkan riyo-shasu-
ga 4500 man [Thank you for always using Twitter for so many years. Thanks to you all,
we have more than 45 million] [Tweet]. Twitter. https://twitter.com/TwitterJP/status/
923671036758958080.
Ueda, K., Nagai, T., Tada, T., Murai, S. & Kishimoto, T. (2017, June 16). “Shin Abe”
ja-narisuto Yamaguchi Noriyuki-shi, reipu giwaku de zettai zetsumei [Rape allegation
pushed “Abe’s close buddy” Journalist Noriyuki Yamaguchi to the brink]. Shu-kan Asahi,
122(30), 24–25.
Yamaguchi, N. (2017, December). Dokusen shuki watashi wo uttaeta Shiori-san he [Exclusive
memoir Dear Shiori who accused me]. Gekkan Hanada, 19, 254–273.
Yoshimura, H., & Ījima, K. (2017, February 9). Kokuyuchi no baikyakugaku hiko-hyo-
[Selling price of government owned land was kept secret]. Asahi Shimbun, Osaka, 37.
Zaimusho-, kakikae mitomeru [MOF admitted forgery]. (2018, March 12). Asahi Shimbun,
Evening edition, 1.
8
TIME TO BE HAPPY
WhatsApp and phatic communication within the
extended family

Noa A. Hatzir

Introduction
Communication between family members is the building block of the family unit, and
the basis for understanding individual experiences as well as sharing values, beliefs, and
perceptions (Galvin et al., 2015). It enables family members to share social reality and
cooperate as a unit (Koerner & Fitzpatrick, 2002). Within the family we learn what
communication is and how to communicate (Bruner, 1990). In recent years there is a
growing research on the extended family due to the increasing life expectancy and its
importance to individuals’ wellbeing (Bengtson, 2001; LaFave & Thomas, 2017; Trilar
et al., 2018), support, care, and treatment (Halperin, 2015).
Although face-to-face communication is quite common among family members,
mediated communication means increasingly make up an important part. In the last
two decades various online communication technologies have emerged, facilitating
and maintaining extended family connections (Chambers, 2017; Miller, 2016;
Seufert et al., 2016). In order to understand how online communication within
families is used, we should therefore look at the perceptions of digital technologies
as well as the cultural and social norms surrounding them (Madianou, 2015;
Madianou & Miller, 2013, 2018; Sinanan, 2016).
The present study focuses on family communication through the Mobile Instant
Messaging (MIM) app, WhatsApp. WhatsApp was chosen to be studied for several
reasons. First, it is the most popular communication application today, mostly used
in America, Asia, the Middle East, and South Africa (Newman et al., 2019; Sevitt,
2017). In Israel 91% of the adult population are using the app (Bezeq, 2020).
Second, communication with WhatsApp is perceived as “natural” (Church & De
Oliveira, 2013), or in other words it is seen as the closest to face-to-face commu-
nication compared to other online platforms, such as emails or Facebook (Chan,
2018). This is mainly due to the versatility of expressions through its interface that

DOI: 10.4324/9781003120100-9
160 N. A. Hatzir

enables synchronous audio or video calls as well as sending asynchronous text,

audio, images, and video. Yet one simple feature makes this channel superior as it
breaks the dichotomy of synchronous–asynchronous communication: the sender
gets notified when the recipient receives and reads the messages.
Most researchers show that WhatsApp helps to maintain the social relationship
(Vanden Abeele et al., 2018). It was found to be a useful and trustful channel that
was increasingly used for intimate conversations with close friends and family, and a
space to share emotions (Alinejad, 2019; Blabst & Diefenbach, 2017; Karapanos et
al., 2016; Sinanan, 2016; Waterloo et al., 2018). Furthermore, it strengthens the
solidarity between family members (Hänninen et al., 2018). On the other hand, it
was argued that WhatsApp encourages phatic communication (Yus, 2017), “small
talk” between two people or more, in which the content itself is less important
than the actual act of communication (Miller, 2008).
In this study the implications of the growing use of WhatsApp and its phatic bias
among family members are discussed, alongside the pros and cons of semantic network
analysis in the field of mediated communication and the way families use MIM. The
chapter includes two case studies: 1) analysis of actual WhatsApp messages among
members of the extended family, and 2) analysis of in-depth interview transcripts on
the perceptions and uses of WhatsApp among three generations of women in a family.

The role of WhatsApp within the family

The variety of communication forms available with WhatsApp make it a useful com-
munication channel among and across all age groups (Chan, 2018; Matassi et al., 2019;
Rosales & Fernández-Ardèvol, 2016). The app automatically uses the personal mobile
phone number as a username and is linked to the list of contacts (Church Rosenfeld
et al., 2018; Seufert et al., 2016). This is one of the reasons that WhatsApp by its
interface design contributes to maintaining strong ties (Chan, 2018).
One of the prominent advantages of WhatsApp (and other MIM) over other
communication channels, such as letters, phone conversation or face-to-face, is the
expansion of synchronous and asynchronous communication features (Karapanos et
al., 2016; Taipale, 2019).
Furthermore, in synchronous communication the discourse is richer. Users interact in
real-time and participants are expected to be available and respond quickly. Together
with the use of voice and video conversations, the ability of WhatsApp users to recognize
if the respondent is available and read the instant messages is in fact a function of instant
confirmation that exists in synchronous communication. Ling and Lai (2016) show that
MIM apps facilitate group arrangements and allow “micro-coordination” (Ling & Lai,
2016). Among family members WhatsApp is found to be useful as a functional means to
coordinate family gatherings, where communication is conducted in a playful and
enjoyable manner (Matassi et al., 2019; Taipale, 2019).
Other researchers emphasize the privacy aspect of the app as a facilitator of intimate
discourse among close friends and family, and as a space that allows people to share
emotions (Blabst & Diefenbach, 2017; Karapanos et al., 2016; Mendoza Pérez &
 Time to be happy 161

Morgade Salgado, 2020; Waterloo et al., 2018) and experiences, and even document
everyday life, a process called “lifelogging” (Karapanos et al., 2016). In addition, What-
sApp has been perceived as a facilitator of social support (Chan, 2018), enhancing the
quality of relationships, and contributing to intimacy, closeness (Knop et al., 2016;
Mendoza Pérez & Morgade Salgado, 2020; Sinanan, 2016), and personal wellbeing
(Kaye & Quinn, 2019). Moreover, WhatsApp is effective in maintaining the connections
between loved ones, in which the sense of “togetherness” can be experienced from a
distance (Aharony & Gazit, 2016; Karapanos et al., 2016; Mendoza Pérez & Morgade
Salgado, 2020; O’Hara et al., 2014; Taipale, 2019). The sense of togetherness goes hand-
in-hand with the principles of phatic communication (Vanden Abeele et al., 2018).
Phatic communication was first defined by Malinowski (1923) as an utterance
whose function is not to inform and share information, but rather to build social
bonds (Malinowski, 1923). Jacobson (1960) developed the linguistic model that
presents six functions of language, including the phatic function. According to him,
this function maintains the connection, physical and phycological, between the
speakers (Jacobson, 1960). Phatic communication includes “small talk”, such as the
exchange of general greetings and metacommentary, which emphasize positive
feelings (Schandorf, 2012). Nowadays, researchers show that in social network
websites and instant massages applications, phatic communication is dramatically
increased (Maíz-Arévalo, 2017; Miller, 2008; Schandorf, 2012; Wang et al., 2011,
2012), particularly in WhatsApp (Wang et al., 2011; Yus, 2017).
Phatic communication, sometimes compared to “grooming” among primates, is
beneficial for social coherence. From the functional perspective, it provides the basic
human need for connection and relatedness (Vanden Abeele et al., 2018; Wang et al.,
2011, 2012). It helps to “establish and maintain social connections together with
opening up to others and recognizing them as potential communication partners”
(Maíz-Arévalo, 2017: 433), and to promote a sense of belonging and community
identification (Parks, 2010). As a result, it promotes social harmony and avoids conflict
(Miller, 2017). Furthermore, phatic communication was found significant in order to
collaborate as a group through mediated communication (Maíz-Arévalo, 2017), and
to replace the nonverbal cues of signaling attention (Kulkarni, 2014). The technology
answers those basic needs (Wang et al., 2011, 2012), and constitutes an extension of
the face-to-face communication (Schandorf, 2012).
From a critical point of view, however, it can be argued that phatic messages
produce “shallow” communication because of the obligation to keep the social
links “alive” through text (Miller, 2008) or photos (Lobinger, 2016). Users are
measured by the number of their friends or followers, and there is a demand to
expand the network (Miller, 2008). Following this view, the sense of being con-
nected has become more critical, sometimes even more than the substantive con-
tent of communication (Miller, 2008, 2017). Yet, it is important to note that
Miller refers to social network websites, characterized by antonymous (forums and
blogs) or pseudo-antonymous (such as Twitter or Facebook) identities, and does
not emphasize the basic human need for connection and relatedness (Vanden
Abeele et al., 2018; Wang et al., 2011, 2012).
162 N. A. Hatzir

The extended family in Israel and its WhatsApp communication

The local cultural context is certainly important in the study of WhatsApp
communication. Taipale (2019) found that in Italy family values are encour-
aged; family members live in proximity and often interact with each other.
WhatsApp was used as a space, especially among the young members of the
family, to conduct long discussions (Taipale, 2019). On the other hand, in
Finland, for example, individual values are emphasized. As the physical distance
between extended family members is often considerable, WhatsApp is used as a
substitute for face-to-face communication (Hänninen et al., 2018; Taipale,
2019). WhatsApp is widely and frequently used in Israel among all age groups,
with 91% of adults using WhatsApp (Bezeq, 2020). It is also customary in Israel
that each person has at least two WhatsApp groups shared by family members,
one for the nuclear and another for the extended family (Aharony & Gazit,
2016).
The current study focuses on Israeli-Jewish families, which at the end of
2018 accounted for 79% of the Israeli population (Central Bureau of Statistics,
2020). In this society, family values are emphasized, and birth rates are higher
compared to other Western countries. Parent–child proximity plays a significant
role (Katz, 2009) and several generations live side by side in the same geo-
graphical area (Lavee & Katz, 2003). Another aspect of Israeli society is the
culture of community, collectivism, and tribalism that is further rooted in the
founding of the state (Almog, 2000). It is characterized by frequent family
gatherings around meals with the extended family members on weekends,
holidays, and birthdays, which contribute to the stability, the sense of con-
tinuity, the preservation of tradition, and the family values (Possick, 2008).
Grandparents take a major part in the lives of parents and grandchildren
(Attar-Schwartz & Buchanan, 2018). Also, it is of the utmost importance for
the older generation to stay connected with the extended family, which pro-
vides meaning and improves their wellbeing. Alongside, there are mixed feel-
ings regarding the expectations of grandparents for communicating with their
grandchildren, which are not always fulfilled. On the other hand, it has been
found that the older generation is cautious not be a burden to their family
members (Halperin, 2015).
In summary, WhatsApp groups of extended families in Israel are a unique sub-
ject of study. The importance of WhatsApp use among the extended family
members, and the tendency to share information provides a rich setting for
research. WhatsApp was also found to encourage phatic communication. There is a
growing academic interest in WhatsApp family groups in recent years, and this
chapter aims at exploring the advantages of semantic network analysis in this field.
The data consisted of two types of texts: (1) analysis of WhatsApp messages among
two groups of extended families (including 30 members in total), and (2) analysis of
the perceptions and uses of WhatsApp from the perspective of three generations of
 Time to be happy 163

women through in-depths interviews (including 30 women—grandmothers,

mothers, and daughters—from ten different families).

Case study 1: Analysis of WhatsApp messages of extended family

groups
To date, researchers who have studied WhatsApp family groups have done so
using research methods such as questionnaires and surveys (Aharony & Gazit,
2016; Blabst & Diefenbach, 2017; Karapanos et al., 2016), qualitative interviews
(Hänninen et al., 2018; Matassi et al., 2019; O’Hara et al., 2014; Taipale, 2019),
ethnographic studies (Alinejad, 2019; Mendoza Pérez & Morgade Salgado, 2020;
Miller, 2016; Sinanan, 2016), and cell phone tracking to determine the location
and time at which calls were made (Rosales & Fernández-Ardèvol, 2016).
The corpus of this analysis is based on the discourse of extended family What-
sApp groups. It should be noted that there is difficulty to reach this kind of data
due to its private nature (Rosenfeld et al., 2018). Barbosa and Milan (2019) discuss
the ethical challenges while conducting ethnographic nature on WhatsApp. They
encourage more research on chat apps as long as they comply with high ethical
standards. For example, research should embrace transparency, avoid dishonest
bypass and guarantee anonymity (Barbosa & Milan, 2019).
To date, researchers who have studied WhatsApp family groups have done so
using research methods such as questionnaires and surveys (Aharony & Gazit,
2016; Blabst & Diefenbach, 2017; Karapanos et al., 2016), qualitative interviews
(Hänninen et al., 2018; Matassi et al., 2019; O’Hara et al., 2014; Taipale,
2019), ethnographic studies (Alinejad, 2019; Mendoza Pérez & Morgade Sal-
gado, 2020; Miller, 2016; Sinanan, 2016), and cell phone tracking to determine
the location and time at which calls were made (Rosales & Fernández-Ardèvol,
2016).
In this study, there is an initial attempt to analyze the actual content of the
messages among family WhatsApp groups. Two WhatsApp groups of extended
families were analyzed with the consent of their members. All the identifiable
details were anonymized. In this analysis, there were two types of “actors”. One
was the family members of the group (users), including 16 users in family A and 14
users in family B. The other was the frequent words that were part of their
WhatsApp discourse. The data were exported from WhatsApp conversations for six
months between December 2018 to June 2019. 1579 messages were sent in group
A and 736 messages in group B during that period.
The process of semantic network analysis included a stage of cleaning up and
removing stop-words (such as “yes”, “no”, “me”, and so on). In addition, a process of
stemming and lemmatization was conducted to include different derivations of similar
words as one entity (for example, “call” and “calling”). Then, we identified the 100
most frequent words, and created a link-list of pairs of words that were mentioned
together in the same message. The intensity of the link between each pair of words
reflects the number of WhatsApp messages in which they were co-mentioned. The
164 N. A. Hatzir

link-list was then imported into the Visone network analysis and visualization software
(Brandes & Wagner, 2004). We ended up with two networks of frequent words, one
for each WhatsApp group. In each network, we conducted cluster analysis using
Louvain modularity and identified the betweenness centrality of the words.

The topics and uses of the extended family WhatsApp groups

Figure 8.1 displays the networks of each group based on the frequent words appearing
together in the same post. In both cases, to allow more readable and comparable net-
works, the figures show only the top 100 links between frequent words. The width of
the links represents the number of posts containing each pair of frequent words. The
size of the nodes represents their relative betweenness centrality. Rectangular nodes
represent the participants. Real names were anonymized based on their family role
(such as grandmother, mother, and daughter). Circle nodes are of the actual words that
were used, and “media” refers to images and videos that they sent.
As can be observed in Figure 8.1, in family A (left) the most prominent and
active member in the network was the grandmother. The most common use of
this group was sharing images, as indicated by the very high centrality of the media.
The strong link between the grandmother and the media indicates that she was also
a very active in sending and commenting on images, with prominent words such as
“beautiful”, “cute”, and “fun”. Another prominent topic discussed in the group
was birthdays and holidays. Here the function of the WhatsApp communication
was mostly greetings and micro-coordination (Ling & Lai, 2016), organizing family
meetings to celebrate the special events as indicated by words such as “tomorrow”,
“morning”, “Friday”, “hour”, “invite”, and “celebrate”.

FIGURE 8.1 Network of frequent words in two WhatsApp groups.

Note. Words are translated to English. The size of the nodes reflects their relative
betweenness level, and the width of the ties the number of sentences mentioned each
pair of words together. Orange nodes reflect the family role, gray nodes the words
themselves.
 Time to be happy 165

Similarly, the discourse in the WhatsApp group of the extended family B (right)
evolved mostly around events. In fact, the greeting “happy birthday” was so dominant
and central in this group that it overshadowed the rest of the network structure. To
understand the patterns and relationships between the other topics, those two words
were removed. In this network grandparents are not active in the network, and the
most central role was played by one of the mothers (labeled as “Mom3”). Prominent
words such as “love”, “health”, “hugs”, and “kisses” were all related to birthday greet-
ings. Another prominent use of WhatsApp, as seen in the previous group, was sharing
photos, as indicated once again by the high centrality of the “media”.
In both WhatsApp groups the most prominent and central words represented
positive experiences such as “happiness” and “fun”. The topics of discussion revolved
mainly around happy events such as the celebration of birthdays and holidays.
Interestingly, in both groups the most prominent emoji that was shared by many
family members was of the kissing face , indicating that one of the important
functions of the group was sharing positive emotions of affection and love.

The positivity bias of family groups in WhatsApp

Looking at the content of WhatsApp conversations, we found mostly positive words
and emojis, reflecting affection and love. In line with Taipale (2019), Matassi et al.
(2019) and Miller (2017), these conversations remained very short and completely
avoided complex or difficult topics, which may have jeopardized the sense of being
together. Expressions of positive emotions in social networks were found to be
“more appropriate”, while negative emotions (Waterloo et al., 2018) and deeper
conversations (Knop et al., 2016) tend to exist in face-to-face encounters (Waterloo
et al., 2018). According to Miller (2016) complex issues and negative emotions can
potentially be discussed in more intimate groups due to the need for privacy. Thus,
in the case of extended family WhatsApp groups there is no room for intimate
communication, resolution of conflicts, and discussions of complex issues.
Supporting the positivity bias, we found that sending media (pictures and videos)
was a significant element of communication between family members. This is a major
way to share happy events with family members and receive positive feedback (Sina-
nan, 2016), and can also be seen as “phatic photo sharing”, which serves immediate
communicative purposes (Lobinger, 2016). Responses reinforce the positive discourse,
with reference mainly to the appearance of the family member in the media and the
situation in which she/he presents. The word “beautiful” frequently appeared in both
groups in response to shared media. Accordingly, Tee and associates (2009) argue that
sharing pictures reflects a commitment among the extended family members, and
often intensifies mutual feelings of connectedness (Tee et al., 2009).
The content analysis also revealed topics that could be unique to the local character-
istics of extended families in Israel. In both WhatsApp groups the discussions pre-
dominantly revolved around rituals such as birthdays, holidays, and weekend greetings.
Family members tended to greet the celebrant with poems and multiple affectionate
words such as: health, happiness, love, and comfort. According to Possick (2008),
166 N. A. Hatzir

birthdays in Israel are central social events that contribute to strengthening family tradi-
tion and culture as well as an opportunity to celebrate and sanctify the value of life.

Case study 2: Analysis of in-depth interviews with three generations

of women
The corpus for the second analysis is based on transcripts of 30 semi-structured
qualitative interviews. Similar to a study by Veronese et al. (2018), we employed
semantic network analyzes of interview transcripts (Veronese et al., 2018). The
qualitative approach helped us to understand the perceptions, find meanings,
interpret, and understand the use of WhatsApp by the users themselves, using their
own language and experiences from their point of view (Denzin & Lincoln, 2000).
The interviews were made with ten Israeli middle-class families, and included three
generations: grandmothers, daughters, and granddaughters. The participation in the
study held two conditions: First, women were adults living separately, and second a
consent was required from interviewees. The interview with each member was held
separately (and not dyadic or triadic) to allow privacy and enable each participant to
describe her personal experience freely (Reczek, 2014). Women were chosen as the
focus for the interviews due to their central and active role in maintaining the contact
with the entire extended family members (Gubernskaya & Treas, 2016; Kamal et al.,
2016; Taipale, 2019). The recruiting process was by snowball sampling.
The interviews were conducted from February to April 2019 and lasted between 35
to 110 minutes. The younger generation ranged from 23 to 28 (M = 26.1), the middle
generation ranged from 47 to 59 (M = 52.9), and the older generation ranged from 70
to 88 (M = 77.7). The interviews dealt with the relationships between the generations
in the family, the pros and cons of the various media channels, including WhatsApp,
and their use in daily life. All interviews were transcribed, and only the sentences
relating to WhatsApp were selected for the present analysis. In total, 447 sentences
from the transcriptions about WhatsApp were analyzed. This study focuses therefore
on the various functions of the WhatsApp groups in the family.

The perceived role of WhatsApp within the extended family

Figure 8.2 portrays the network of frequent words mentioned together with the
word “WhatsApp” by the interviewees. To make the network readable, the figure
includes only pairs of words that appeared together in at least three sentences or
more. It displays four clusters of words based on Louvain modularity as well as the
most central words based on their betweenness centrality. The four clusters
obtained from the analysis represent three distinctive themes that highlight the role
of WhatsApp communication within the extended family.
Cluster A on the left describes the communication forms available through WhatsApp
and its advantages over other platforms and channels such as face-to-face, phone,
email, and Facebook. The main advantages described by the interviewees were the
ability to connect simultaneously with the entire extended family. Other prominent
features raised were the flexibility to choose between diverse communication forms,
 Time to be happy 167

FIGURE 8.2 Network of frequent words in interview transcripts on WhatsApp in the

family.
Note. Words are translated into English. The size of the nodes reflects their relative
betweenness level, and the width of the ties the number of sentences mentioned each
pair of words together.

including video and pictures, and the ease of using the WhatsApp’s interface. One of
the interviewees from the second generation (mothers), explained why she prefers to
share the good news through her WhatsApp family group and not by other platforms:

Because with WhatsApp I can post it [the good news] directly to everyone,
because I will not start making 20 phone calls.
(Mother, 53 years)

Another woman explained that she likes to celebrate good news with the family, parti-
cularly when she believes that it will be interesting for them. She first writes or sends
photos on WhatsApp, and only then talks about it on the phone with her family members:

I have diabetes, and because I started a new diet the sugar went down won-
derfully, so I told everyone about that. They were all very happy. And now
they ask me how my sugar rates are, every day.INTERVIEWER: All this happens
through WhatsApp?
168 N. A. Hatzir

Yes.You see … it is a good thing. If someone gives birth, so first—WhatsApp,

later we speak and send pictures of the newborn baby …
(Grandmother, 70 years)

Cluster B on the right evolves around the word “know”, which connects the words
“mother” and “daughter”. The main idea presented by the interviewees was that the
usage of WhatsApp depends on the quality of the personal relationship between the
family members, and particularly the mother–daughter relationship. Knowing their
daughter’s personality and schedule, mothers choose how and when to use different
types of communication channels. They also emphasized the importance of WhatsApp
in maintaining good relationships, particularly between grandparents and their grand-
children, or when the children move away from their parent’s home.
Considering the personal relationship, one of the women said that the nice part of
WhatsApp is that it helps her and her daughter to feel close, although they live apart.

Really, there is a desire to feel that we are … even if we are far from each
other, we’re still close. I will even send her pictures of things I want to buy so
that we can feel together on a daily basis.
(Mother, 49 years)

Most women mentioned that they would prefer phone conversations over
WhatsApp messages, but this is not always possible. Sending WhatsApp messages
allows them to communicate with minimal interference:

During the day, I know she is busy, she told me that. I would like to make a
phone call, but she told me that she is in meetings, and can’t answer. So, it is
better to send her a WhatsApp [message] and she will see it later.
(Grandmother, 76 years)

The fear of disturbing was mentioned in other contexts. One mother explained
that she was mindful of her daughter’s moods, and respected her preference for
communicating through WhatsApp rather than over the phone:

Sometimes she prefers to chat and not to talk. It’s like she does not feel like talking.
Especially if she is depressed about something, she prefers to write and not talk. I go
with it. Even if I call her, she won’t pick up, so I know that and therefore send her a
message. I know she is not in the mood for talking; she just doesn’t feel like it.
(Mother, 58 years)

Finally, clusters C and D on the top and bottom are related to the time dimension, with
words such as “time”, “available”, and “convenient”. This theme is related to the ability
(with the keywords “able”, “available”), and willingness (with the keywords “want” and
“prefer”) of the family members to engage in synchronous communication such as
talking and making phone calls or asynchronous communication: writing and sending
 Time to be happy 169

messages. The variety of features available through the WhatsApp interface allows both
synchronous and asynchronous communication. Yet respondents often empha-
sized its semi-a-synchronous qualities, such as the ability to send a message, see
whether someone saw it, and learn about her/his availability and willingness before
committing to time-consuming synchronous conversation over the phone.
One mother emphasized the importance of her independence after all her children
left the house. WhatsApp enabled her to put limits and determine her availability.

She wants something; I am not available right now to hear what she wants.
She writes to me; I can only send her an icon, which symbolizes—I under-
stand, I am on it. Just not at this moment, because now I study.
(Mother, 59 years)

A young interviewee explained that she is often very busy. She expressed frus-
tration about her family’s expectation to communicate immediately, and said that
she tends to answer a few days after everyone:

They [the family members] already know. If I send someone “happy birthday”
on the accurate day, it is as if something happened. They already know I am
not available. I just do everything retrospectively. Sometimes I am in the
middle of something, like work, and then I forget … the conversation on
WhatsApp is endless, sometimes I prefer to just call and finish with it.
(Daughter, 28 years)

The perceived strengths of WhatsApp in the family

The interviews emphasized the major advantages of WhatsApp as a communication
channel within extended family communication. The first is its ability to connect
“everyone” and share good news. In line with previous studies, the findings show that
WhatsApp allows family members to update each other on a daily basis and share
important events (Karapanos et el., 2016; Rosales & Fernández-Ardèvol, 2016; Sina-
nan, 2016). A second prominent feature mentioned was the flexibility to choose
between diverse communication forms, including video and pictures. Another
advantage mentioned was the flexibility to send messages at any time and place and to
control the number of recipients. This adjustability of WhatsApp allows inter- and
cross-generation communication and therefore can bridge digital gaps (Chan, 2018;
Matassi et al., 2019; Rosales & Fernández-Ardèvol, 2016; Taipale, 2019).
WhatsApp has become a prime communication channel between mothers and
daughters. One perceived benefit of WhatsApp was the ability to send private messages
and not only within the family group. Users can feel connected even if someone is not
immediately available. Finally, WhatsApp was found to be useful when dealing with
tension and conflicts, and the ability to indirectly express negative emotions such as
anger and frustration. In those cases, WhatsApp was perceived as means to avoid direct
conversation but at the same time keeping the feeling of connectedness.
170 N. A. Hatzir

Discussion
Due to its diversity and unique features, WhatsApp has become a crucial part of
our media environment. It has been embraced as one of the prime channels for
communication within the family around the world (Madianou, 2015; Madianou
& Miller, 2013, 2018; Sinanan, 2016), and particularly among family members in
Israel (Bezeq, 2020). In this study we looked at the perception and actual uses of
WhatsApp groups among the extended family in Israel. The findings highlight the
perceived benefits of WhatsApp to maintain the relationships among the extended
family members thanks to its versatility, allowing the use of different communica-
tion forms (text, audio, image, and video), and more importantly the integration of
synchronous and asynchronous communication.
We utilized semantic network analysis of two corpora: actual WhatsApp con-
versations and interview transcripts. This enabled us to present two complementary
approaches to the understanding of the relationship between technology and its
users. On the one hand, we showed how the unique features of WhatsApp con-
tributed to the communication among the extended family communication. Its
perceived benefits are evident: a variety of synchronous and asynchronous forms of
communication, the transmission of messages in a variety of ways (e.g., text,
image, audio, video), and the ability to create groups that include a large number
of participants, free of charge based on the contact persons stored on the mobile
phone. In practice, family members use WhatsApp mainly to send positive mes-
sages and share images of daily events that contribute to strengthening the sense
of “unity” (Aharony & Gazit, 2016; Karapanos et al., 2016; O’Hara et al., 2014;
Taipale, 2019).
When analyzing the content of conversations within the extended family, we found
mainly a positivity bias around photo sharing and family rituals (birthdays, holidays,
and weekend greetings or gatherings), where content involved fun experiences and
affectionate feelings. This provides evidence for the dominance of phatic commu-
nication in WhatsApp. Supporting this view, many of the interviewees claimed that
WhatsApp is less suitable for serious interpersonal communication, intimate discus-
sions, and resolution of conflicts. They indicated that in those cases a phone con-
versation is preferable, as it allows a higher degree of synchrony and involvement, and
more privacy. Yet with the growing use of WhatsApp and other mediated means of
communication within the family, it may be that phatic communication will become
the norm, as people are increasingly busy in being connected and maintaining their
positive image (Miller, 2008, 2017).

Limitations and future directions

Due to privacy constraints and difficulty accessing this type of data (Barbosa &
Milan, 2019; Rosenfeld et al., 2018), only two WhatsApp groups were studied in
the first case study. Both groups similarly displayed the importance of rituals in
Israeli society, such as greetings on weekends, holidays, and birthdays. In future
 Time to be happy 171

studies we recommend extending this investigation to examining how social and

cultural values are expressed in WhatsApp conversations in other countries.
Additionally, the interviews were conducted among families who indicated that
their communication and family relationships were normal or generally good. This
was reflected in the generally positive perceptions toward WhatsApp, and the
positivity bias in the content analysis. Future studies should explore the use of
WhatsApp in families dealing specifically with conflicts and crises. This will provide
a complementary picture of the role of WhatsApp and particularly the strengths
and weaknesses of its unique synchronous and asynchronous features.

Tips and lessons for the use of semantic network analysis

1. The decision on which words to include in the semantic network analysis

depends on the type of corpus in question. When conducting an analysis of
instant massages, it is important to decide whether to differentiate between
the participants involved and the content of the conversation. In this case,
visually differentiating in the network between the users and the frequent
words of their posts helped to assess their unique role in the family. For
example, it was possible to identify who is dominant and active in the group,
and what type of content is produced by each family member.
2. When analyzing interview transcripts, it is important to keep the research
question in mind and choose the relevant words for the specific research. In
this paper, the interviews were dealing with all sorts of platforms, but only the
sentences with the word “WhatsApp” were included, enabling them to focus
and offer a graphic representation of the discourse on WhatsApp.
3. A meaningful interpretation of the networks requires a good knowledge of the
content. When looking at the network of interview transcripts, it was particu-
larly helpful to read the interviews back and forth to build coherent themes and
understand their broader context. Semantic network analysis does not replace
qualitative analysis; rather, it can assist in identifying some of the main themes
and organizing the data as a complementary way to look at the themes.
4. Semantic network analysis is particularly useful when the text is large
enough. When using shorter texts, such as in the case of the interview
transcripts that included only 447 sentences that mentioned the word
“WhatsApp”, the network is relatively small and simple. This is still possible
but limits the ability to make more general claims.
5. Finally, similar to other studies presented in this book, even when it comes to
shorter texts, there is a need to sparsity the network. For example, in the first
case study, the word “birthday” was dominant in the massages of family B. As
we know that most conversations in this group were related to birthdays;
deleting these words enabled us to see more clearly the network structure and
identify the clusters and main themes discussed in the context of birthdays.
172 N. A. Hatzir

References
Aharony, N., & Gazit, T. (2016). The importance of the Whatsapp family group: an
exploratory analysis. Aslib Journal of Information Management, 68(2), 174–192.
Alinejad, D. (2019). Careful Co-presence: The Transnational Mediation of Emotional Inti-
macy. Social Media + Society, 5(2), 1–15.
Almog, O. (2000). The Sabra: the Creation of the New Jew. Berkeley, CA: University of California
Press.
Attar-Schwartz, S., & Buchanan, A. (2018). Grandparenting and adolescent wellbeing: evi-
dence from the UK and Israel. Contemporary Social Science, 13(2), 219–231.
Barbosa, S., & Milan, S. (2019). Do Not Harm in Private Chat Apps: Ethical Issues for Research
on and with WhatsApp. Westminster Papers in Communication and Culture, 14(1), 49–65.
Bezeq. (2020). The Digital life: Bezeq internet report 2019–2020. https://media.bezeq.co.
il/pdf/internetreport_2019.pdf.
Bengtson, V. L. (2001). Beyond the nuclear family: the increasing importance of multi-
generational bonds: the burgess award lecture. Journal of Marriage and Family, 63(1), 1–16.
Blabst, N., & Diefenbach, S. (2017). WhatsApp and wellbeing: A study on WhatsApp usage,
communication quality and stress. HCI 2017: Digital Make Believe—Proceedings of the 31st
International BCS Human Computer Interaction Conference, HCI 2017, July 2017, 1–6.
Brandes, U., & Wagner, D. (2004). Visone: Analysis and Visualization of Social Networks.
In M. Jünger & P. Mutzel (Eds.), Graph Drawing Software (pp. 321–340). New York:
Springer-Verlag.
Bruner, J. (1990). Acts of Meaning. Cambridge, MA: Harvard University Press.
Central Bureau of Statistics. (2020). Family Day—families and households in Israel. https://
www.cbs.gov.il/he/mediarelease/DocLib/2020/053/11_20_053b.pdf.
Chambers, D. (2017). Networked intimacy: Algorithmic friendship and scalable sociality.
European Journal of Communication, 32(1), 26–36.
Chan, M. (2018). Mobile-mediated multimodal communications, relationship quality and
subjective wellbeing: An analysis of smartphone use from a life course perspective. Com-
puters in Human Behavior, 87, 254–262.
Church, K., & De Oliveira, R. (2013). What’s up with WhatsApp? Comparing mobile instant
messaging behaviors with traditional SMS. MobileHCI 2013—Proceedings of the 15th
International Conference on Human-Computer Interaction with Mobile Devices and
Services, August 21–30, Munich, 352–361.
Denzin, N. K., & Lincoln, Y. S. (Eds.) (2000). Introduction: The Discipline and Practice of
Qualitative Research. New York: SAGE Publications.
Galvin, K. M., Braithwaite, D. O., & Bylund, C. L. (2015). Family Communication: Cohesion
and Change (9th ed.). London: Routledge.
Gubernskaya, Z., & Treas, J. (2016). Call Home? Mobile Phones and Contacts With
Mother in 24 Countries. Journal of Marriage and Family, 78(5), 1237–1249.
Halperin, D. (2015). Intergenerational relations: The views of older Jews and Arabs. Journal
of Intergenerational Relationships, 13(1), 51–74.
Hänninen, R., Taipale, S., & Korhonen, A. (2018). Refamilisation in the broadband
society—the effects of ICTs on family solidarity in Finland. Journal of Family Studies, 1–17.
Jacobson, R. (1960). Linguistics and poetics. In T. Sebeok (Ed.), Style in Language (pp. 350–377).
Cambridge, MA: MIT Press.
Kamal, F. M., Noor, N. L. M., & Baharin, H. (2016). “Silence is golden no more” in family digital
environment: Understanding the kinkeeper role through mobile social messaging system research. 24th
European Conference on Information Systems, ECIS 2016. June 12–15, Istanbul.
 Time to be happy 173

Karapanos, E., Teixeira, P., & Gouveia, R. (2016). Need fulfillment and experiences on social
media: A case on Facebook and WhatsApp. Computers in Human Behavior, 55, 888–897.
Katz, R. (2009). Intergenerational family relations and subjective wellbeing in old age: A
cross-national study. European Journal of Ageing, 6(2), 79–90.
Kaye, L. K., & Quinn, S. (2019). Psychosocial Outcomes Associated with Engagement
with Online Chat Systems. International Journal of Human-Computer Interaction, 36(2),
190–198.
Knop, K., Öncü, J. S., Penzel, J., Abele, T. S., Brunner, T., Vorderer, P., & Wessler, H. (2016).
Offline time is quality time. Comparing within-group self-disclosure in mobile messaging
applications and face-to-face interactions. Computers in Human Behavior, 55, 1076–1084.
Koerner, F. A., & Fitzpatrick, M. A. (2002). Understanding Family Communication Patterns
and Family Functioning: The Roles of Conversation Orientation and Conformity
Orientation. Annals of the International Communication Association, 26(1), 36–65.
Kulkarni, D. (2014). Exploring Jakobson’s “phatic function” in instant messaging interac-
tions. Discourse and Communication, 8(2), 117–136.
LaFave, D., & Thomas, D. (2017). Extended families and child wellbeing. Journal of Devel-
opment Economics, 126, 52–65.
Lavee, Y., & Katz, R. (2003). The Family in Israel: Between Tradition and Modernity.
Marriage & Family Review, 35(1–2),193–217.
Ling, R., & Lai, C. H. (2016). Microcoordination 2.0: Social Coordination in the Age of
Smartphones and Messaging Apps. Journal of Communication, 66(5), 834–856.
Lobinger, K. (2016). Photographs as things—photographs of things. A texto-material per-
spective on photo-sharing practices. Information Communication and Society, 19(4), 475–488.
Madianou, M. (2015). Polymedia and Ethnography: Understanding the Social in Social
Media. Social Media and Society, 1(1).
Madianou, M., & Miller, D. (2013). Polymedia: Towards a new theory of digital media in
interpersonal communication. International Journal of Cultural Studies, 16(2), 169–187.
Madianou, M., & Miller, D. (2018). Polymedia: Towards a new theory of digital media in
interpersonal communication. The Monitoring of Public Opinion: Economic and Social Changes
Journal, 1(143), 334–356.
Maíz-Arévalo, C. (2017). “Small talk is not cheap”: phatic computer-mediated commu-
nication in intercultural classes. Computer Assisted Language Learning, 30(5), 432–446.
Malinowski, B. (1923). The problem of meaning in primitive languages. In C. K. Ogden &
I. A. Richards (Eds.), The Meaning of Meaning: A Study of the Influence of Language upon
Thought and of the Science of Symbolism (pp. 296–336). London: Routledge & Kegan Paul.
Matassi, M., Boczkowski, P. J., & Mitchelstein, E. (2019). Domesticating WhatsApp: Family,
friends, work, and study in everyday communication. New Media & Society, 21(10), 2183–2200.
Mendoza Pérez, K., & Morgade Salgado, M. (2020). Mobility and the mobile: A study of
adolescent migrants and their use of the mobile phone. Mobile Media and Communication, 8
(1), 104–123.
Miller, D. (2016). What is social media? In D. Miller, E. Costa, N. Haynes, T. McDonald,
R. Nicolescu, J. Sinanan, J. Spyer, S. Venkatraman, & X. Wang (Eds.), How the World
Changed Social Media (pp. 19–25). London: UCL Press.
Miller, V. (2008). New media, networking and phatic culture. Convergence, 14(4), 387–400.
Miller, V. (2017). Phatic culture and the status quo: Reconsidering the purpose of social
media activism. Convergence, 23(3), 251–269.
Newman, N., Fletcher, R., Kalogeropoulos, A., & Kleis Nielsen, R. (2019). Reuters Insti-
tute Digital News Report 2019. https://reutersinstitute.politics.ox.ac.uk/sites/default/
files/inline-files/DNR_2019_FINAL.pdf.
174 N. A. Hatzir

O’Hara, K., Massimi, M., Harper, R., Rubens, S., & Morris, J. (2014). Everyday dwelling with
WhatsApp. Proceedings of the ACM Conference on Computer Supported Cooperative
Work, CSCW, March 14–18, Vancouver, 1131–1143.
Parks, M. R. (2010). Social network sites as virtual communities. In Z. Papacharissi (Ed.), A
Networked Self. Identity, Community, and Culture on Social Network Sites (Issue 2010, pp.
105–123). New York: Taylor & Francis Group.
Possick, C. (2008). The family meal: An exploration of normative and therapeutic ritual
from an ethnic perspective. Journal of Family Psychotherapy, 19(3), 259–276.
Reczek, C. (2014). Conducting a multi family member interview study. Family Process, 53
(2), 318–335.
Rosales, A., & Fernández-Ardèvol, M. (2016). Beyond WhatsApp: Older people and
smartphones. Romanian Journal of Communication and Public Relations, 18(1), 27–47.
Rosenfeld, A., Sina, S., Sarne, D., Avidov, O., & Kraus, S. (2018). A Study of WhatsApp Usage
Patterns and Prediction Models without Message Content. 1–24. http://arxiv.org/abs/1802.03393.
Schandorf, M. (2012). Mediated gesture: Paralinguistic communication and phatic text.
Convergence, 19(3), 319–344.
Seufert, M., Hoβfeld, T., Schwind, A., Burger, V., & Tran-Gia, P. (2016). Group-based
communication in WhatsApp. 2016 IFIP Networking Conference (IFIP Networking) and
Workshops, May 17–19, Vienna, 536–541.
Sevitt, D. (2017). The most popular messaging apps by country. https://www.similarweb.
com/corp/blog/popular-messaging-apps-by-country/.
Sinanan, J. (2016). Social media and sorting out family relationships. In S. Y. Tettegah (Ed.),
Emotions, Technology, and Social Media (pp. 63–82). Amsterdam: Elsevier.
Taipale, S. (2019). Intergenerational Connections in Digital Families. New York: Springer.
Tee, K., Brush, A. J. B., & Inkpen, K. M. (2009). Exploring communication and sharing
between extended families. International Journal of Human Computer Studies, 67(2), 128–138.
Trilar, J., Kos, A., Jazbinšek, S., Jensterle, L., & Duh, E. S. (2018). ICT to promote well-
being within families. Sensors, 18(9), 1–17.
Vanden Abeele, M., De Wolf, R., & Ling, R. (2018). Mobile media and social space: How
anytime, anyplace connectivity structures everyday life. Media and Communication, 6(2), 5–14.
Veronese, G., Pepe, A., Dagdukee, J., & Yaghi, S. (2018). Teaching in conflict settings:
Dimensions of subjective wellbeing in Arab teachers living in Israel and Palestine. Inter-
national Journal of Educational Development, 61(December 2017), 16–26.
Wang, V., Tucker, J. V., & Haines, K. (2012). Phatic technologies in modern society.
Technology in Society, 34(1), 84–93.
Wang, V., Tucker, J. V., & Rihll, T. E. (2011). On phatic technologies for creating and
maintaining human relationships. Technology in Society, 33(1–2),44–51.
Waterloo, S. F., Baumgartner, S. E., Peter, J., & Valkenburg, P. M. (2018). Norms of online
expressions of emotion: Comparing Facebook, Twitter, Instagram, and WhatsApp. New
Media & Society, 20(5), 1813–1831.
Yus, F. (2017). Contextual constraints and non-propositional effects in WhatsApp commu-
nication. Journal of Pragmatics, 114, 66–86.
9
SCHOOL IMPROVEMENT
Semantic network analysis of the literature

Audrey Addi-Raccah, Tali Shahrabani and Elad Segev

Introduction
School Improvement (SI) is crucial for policymakers with different educational
systems across countries (OECD, 2019; Greatbatch & Tate, 2019). In many
Western countries SI is associated with neo-liberal policies (Wrigley, 2008) that
emphasize decentralization (e.g., Nurkolis & Sulisworo, 2018) and embrace a
quasi-market logic in the educational systems by opening schools to competition,
privatization, and accountability. These measures demand educators strive to find
and employ practices that are significant for improving students’ achievements
(Azkiyah, 2017; Hallinger, 2018; Pashiardis & Brauckmann, 2019). It should be
noted that these efforts have been ongoing since the well-known Coleman Report
(Coleman et al., 1966). Nowadays the research literature on SI is highly prolific,
comprehensive, and addresses various actors (e.g., school principals, teachers,
students, policymakers), processes, and outcomes situated within diverse social
and organizational contexts.
The diverse literature reviews in the field (Kovačević & Hallinger, 2019, biblio-
metric analysis; Gumus et al., 2018; Tian & Huber, 2019, bibliometric and content
analyses) show that SI is one of the themes that has preoccupied research. Yet the
focus of these studies has been mainly in the forms of leadership in schools. In the
present study, we focus specifically on SI, attempting to identify and map the general
themes in the current literature (during the five years between 2014 and 2018) that
cover issues beyond leadership. Particularly, we look at whether the current literature
addresses the need for reforms and calls for educational response to the changes of
the 21st century, including the growing globalization and international competition,
and the development of digital technologies (Cheng, 2011).
In this chapter, we first introduce the conceptual relationship between school
effectiveness (SE) and SI. As the literature in the field is extensive, we further

DOI: 10.4324/9781003120100-10
176 A. Addi-Raccah, T. Shahrabani and E. Segev

employ the semantic network analysis to identify reoccurring themes and topics
related to SI as emerging in current research articles. Semantic network analysis
(Segev, 2020) enables to map the relationships and identify patterns within the
knowledge regarding SI in scientific publications. We collected 215 research
articles that mentioned both terms (SE and SI) from the last five years, produced
a list of relevant frequent words, and conducted semantic network analysis to
identify the main clusters and themes discussed in the literature. While semantic
network analysis enables us to map the overall topics and their relationships, we
further “dive-in” and employ a thematic qualitative analysis of a sample articles to
better understand the context and inter-connectedness between topics. To con-
clude, we identify gaps in the literature and suggest some directions for future
research.

The concepts of school effectiveness and improvement

School improvement frequently comes along with school effectiveness. These
two concepts have been extensively discussed and examined in the research litera-
ture (e.g., Stoll & Fink, 1994). Broadly speaking, school effectiveness refers to the
school, classroom, and student factors that are empirically proven to be related to
achievements (e.g., Creemers et al., 2000). More precisely, school effectiveness is
defined as understanding educational processes, and explaining their outcomes in
terms of concrete causes and effects (Hopkins et al., 1994) or as the means–end
relationship between educational processes and students’ outcomes as knowledge
and skills in several domains (Creemers et al., 2000). As for school improvement, it
focuses mainly on school change. According to Louis et al. (1999), improvement is
interchangeable with implementation, reform, or change. They indicated that these
terms reflect “altering the behavior of school employees or the performance of the
school on any set of pre- or post-determined indicators …” (p. 251). School
improvement emphasizes the practices that could be modified and changed for
contributing to higher school achievements (e.g., teachers’ instruction and train-
ing). School improvement and school effectiveness are related, with improvement
being part of the effort for schools’ effectiveness (Hopkins et al., 2014) and rooted
in the essential components of effective schools (Preston et al., 2017). According to
Reynolds et al. (2014) “school effectiveness researchers [are] in close intellectual
proximity to school improvement researchers and practitioners”. They point to the
benefit of merging the two approaches as

many SE researchers became more comfortable with SI’s typical qualitative

methodology, its commitment to more cultural views of school processes
instead of the formal organizational factors that had been the SE commitment,
and its commitment to the importance of seeing teachers as something other
than mere “empirical/rational” educational actors
(p. 199)
 School improvement 177

The overtime landscape in SE and SI research

Several scholars tracked the changes that occurred in the field of SI, while
emphasizing different stages in its development since the mid-20th century (e.g.,
Cheng, 2011, three waves; Hargreave, 2012, four ways; or Hopkins et al., 2014,
five phases). Within this context, Hargreaves (2012) pointed out that from World
War II until the 1970s, professionals within the educational sector led various
innovations and new social initiatives that were informed by intuition and ideol-
ogy. More systematic research has been conducted within the framework of school
effects’ research beginning with the Coleman’s Report (Coleman et al., 1966) that
looked for components, factors, and practices that decoded the “black box” of
school characteristics and internal processes that were associated with SI (Rey-
nolds et al., 2014). Further on, during the 1980s and 1990s, there was a shift from
centralized school systems and centric school approaches to decentralized school
systems and wide efforts for improvement (Hopkins et al., 2014). In the school
centric approach, school principals focused mainly on internal school processes
and actions for achieving planned goals and tasks in learning, teaching, and
schooling. Accordingly, the research examined the correlations between school
organizational conditions and the quality of education they provided to students. Less
consideration was given to the diversities and the influences of the external envir-
onment and stakeholders. In contrast, the move to system-level efforts for improve-
ment was associated with the decentralization of the educational systems,
incorporation of market principles in education, greater competition among schools,
and more pressure for high standards accountability demands (Hargreaves, 2012;
Cheng, 2011; Fullan, 2009).
As presented by Hargreaves (2012), schools are turning into a setting for collec-
tive actions, such as building capacity, shared leadership, enhanced professional
development, and cultivation of school culture and climate that support teachers’
work and students’ learning and outcomes (e.g., Higham & Booth, 2018). In this
phase, the school is the unit of change and improvement, along with the multiple
dimensional contexts in which it is situated (see also Hallinger, 2018), including
exposure to global trends of standards (e.g., knowledge based on PISA, TIMMS)
that enhance competition between nations and educational systems. For improving
schools, studies showed that school leaders were considered as pivotal figures for
improving schools through diverse types of leadership, such as instructional, trans-
formational, or distributive leadership (Leithwood et al., 1999; Spillane, 2012). In
addition, teachers’ instruction and teaching quality were found to have a dominant
and direct influence on students’ outcomes compared to other factors (Hopkins et
al., 2014; Azkiyah, 2017).
Moving into the 21st century, according to Cheng (2011), schools are exposed
to the impact of rapid globalization, information technology, and international
economic and social competition. Hence, current effectiveness and improvement
trends of education, enhance the need to maximize the learning outcomes of the
younger generation, and prepare them for the growing demands of the 21st
178 A. Addi-Raccah, T. Shahrabani and E. Segev

century. Thus, there was a paradigm shift in learning and teaching to ensure their
relevance for the future in the context of multiple intelligence, globalization,
localization, and individualization. Further, schools are perceived not as indepen-
dent units (Keddie, 2014), but rather as part of a broader education system. They
are required to develop a more flexible and heterarchical structure based on
establishing extended social networks and collaboration with diverse internal and
external school stakeholders (e.g., parents, NGOs, teachers, Local Educational
Authorities, students, school networks) (Addi-Raccah et al., 2018; Hallinger, 2018;
Ni et al., 2017) in order to maximize learning opportunities for students (Hopkins
et al., 2014; Cheng, 2011).
This leads to new modes of governance and bottom-up perceptions that focus
on school processes through teachers’ professional learning communities, lateral
collaboration within schools, and networking (Hargreaves, 2012). As indicated by
Muijs (2010), there is an association between networking and collaboration for
school improvement. Apparently, at the present time the field of SI is situated in
multi-dimensional contexts, attempting to facilitate collaboration and networking
within schools and between schools and other institutions, and navigating between
the axis of local and global trends and between bottom-up and top-down approa-
ches. Further, while the research focused mainly on students’ achievement, there is
a consensus upon the need to measure diverse educational outcomes (e.g., the
dynamic model). This direction is further enhanced with the emphasis on 21st-
century skills, and the efforts to advance soft skills and socio-emotional learning
(Chernyshenko et al., 2018). This goes along with Hill and Guthrie’s (1999) notion
that on the eve of the 21st century, there is a need to develop a new school
paradigm that is based on school productivity and improvement (Louis et al.,
1999). If so, in view of the current trends of SI, our aims are to explore what is the
narrative emerging from the current SI literature by revealing the central players
(actors) who are involved in the process of SI and the domains and themes that
underline SI research.

Methods

Sample
Our analysis is based on peer-reviewed articles. The process of obtaining the
sample of articles included several stages. First, we used Clarivate Analytics JCR
journal citation reports to obtain a list of all the journals in the field of education
with any impact factors. Out of all these journals, we chose only those in the top
two quarters of the list. Second, we defined relevant search queries that could be
used to retrieve articles that delt with school improvement. The chosen search
queries that yielded relevant results were: “effective schools” or “effective school
research” or (school and success), “School effectiveness”, and “school improve-
ment”. Searches were conducted in four databases (PsycNET, Education Source,
ERIC, and Web of Science) in the fields of title, summary, and keywords for each
 School improvement 179

of the chosen journals that we defined for the five years between 2014 and 2018.
To conclude, we manually scanned all the articles for relevancy. Out of more than
300 articles retrieved, we omitted those focusing on special education schools,
psychological learning disorders, physical disabilities, and duplication of articles. In
total, this search process returned 215 relevant articles that explicitly delt with
school improvement.

Procedure
Our analysis was conducted in two stages. We employed semantic network analysis
to identify the main themes around which the academic discussion on success in
schools revolved. Once we identified central terms and clusters of words related to
the field, we conducted a qualitative analysis to better understand the context in
which these words and themes were discussed in the literature.
For the semantic network analysis, we first constructed a list of the most fre-
quent words in the text with a frequency of at least 0.01% of the total number of
words (N = 2,299,185). We then manually inspected this list, removed stop-words
and irrelevant words, and merged words with a similar derivation (such as singular
and plural). We ended up with a list of 229 frequent words that appeared at least
139 times in the corpus. We also identified frequent phrases in the literature (such
as “professional development” and “transformational leadership”), which were
crucial in the context of our study. Next, we constructed an undirected weighted
network of words that co-occurred in the same sentence. We also explored the
option of looking at words appearing together in the same paragraph but found
that the window of one sentence was more accurate in our corpus, keeping the
tight proximity between words (see also Danowski, 1993). Finally, we conducted
several analyses, including measuring the network centrality of words, a cluster
analysis to extract the main themes and topics, and, based on the semantic map, an
in-depth qualitative analysis to better understand the context of SI research.
For the following analysis, due to the high density of the network, we focused
on two network resolutions: a network of word pairs that were mentioned in at
least 200 sentences (for visualizing the network’s overall structure and main themes
in Figures 9.1), and another network of word pairs that appeared in at least 100
sentences (around one of the particular keywords in Figures 9.2). In addition,
articles that were selected for further qualitative analysis were based on a propor-
tional stratified sampling of about a quarter (23%; N = 50) of the articles published
in each year (2014 to 2018) (e.g., for an example of a random selection of articles
in conducting a literature review, see McInerney et al., 2004). These articles were
randomly selected from all the articles that emerged from our search in order to
provide a diverse scope of articles, and not necessarily an influential one (prior
reviews conducted content analyses of the most cited articles; Tian & Huber,
2019). For each article, two of the researchers conducted a content analysis gen-
erated by the following codes: research main topic or aims, participants, research
methods, main findings, and conclusion based on the findings. For the current
180 A. Addi-Raccah, T. Shahrabani and E. Segev

study that deploys the dominant players and themes emerging in the current SI
literature, we concentrated on two codings: the participants in each study and its
central topic or issue.

Results
Figure 9.1 presents the results of a cluster analysis based on Louvain’s modularity
(Blondel et al., 2008, see also Chapter 1). The size of the nodes reflects their
betweenness centrality. The width of the ties reflects the number of sentences in
which each pair of words was mentioned.
Figure 9.1 portrays four clusters that revolve around four central words: on the
left “teachers”, on the right “students”, and in between “improve” (bottom) and
“leadership” (top). In more detail, Figure 9.1 shows that the articles dealing with
school improvement emphasize three significant players. The word “teacher”,
which is prominent, is surrounded by prominent words related to professional
development and prospective learning, suggesting that SI involves investment in
human resources in the workplace. The word “student” is surrounded by words that
focus mainly on educational products and academic achievements (such as “tests”,
“scores”, and “assessments”). The word “leaders” focus on three dimensions of lea-
dership: “distributive”, “transformational”, and “instructional” leadership. As this
cluster is located in between the teacher and student clusters, it seems that leaders and

FIGURE 9.1 Semantic network of words related to SI and SE.

Note. The network resolution is 200 sentences and above. Network visualization and
calculations were performed with Visone (Brandes & Wagner, 2004).
 School improvement 181

managers are responsible for enabling the work of teachers and students, and more
generally to SI. Moreover, the three players connect to the cluster of school
improvement that includes the words “capacity”, “networks”, and “policies”.
The link of all these words to improvement is not surprising as our search was
initially focused on school improvement as a keyword. Yet, although the initial
search included the word “effective”, it was more marginal in the network. Hence,
the research literature seems to emphasize the school processes—of improvement—
rather than its outcomes. Moreover, the narrative that emerges from the analysis
revealed two important foci: one is the triangle between teachers, school leaders,
and school improvement. The other is on the triangle of teachers, students, and
school improvement. Leadership seems to be indirectly linked to students through
teachers and through school improvement processes. Interestingly, the parents,
who are significant actors in the field of education, appeared to be minor in the SI/
SE literature, as parents are mostly mentioned in relation to the students, with no
prominent ties to the teachers or school leadership.
Based on our sample of 50 articles, we found that there was indeed a centrality to
teachers (as mentioned above); 27 articles examined teachers or teachers with other
participants (see below for more details). Further, students were dominant in Figure
9.1 and most of the articles addressed the implications of the educational process for
improving students’ outcomes, but in practice only five articles examined students as
their primary subject. Four of these studies referred to efforts and programs for
improving students’ achievements (in math or literacy) (Gorard et al., 2015, 2017;
Kelcey & Shen, 2016, Valenzuela et al., 2016; Bellei et al., 2016), and one for
decreasing school bullying (Kyriakides et al., 2014). It was also found that seven of the
articles addressed schools’ leadership specifically. These studies reflected a wide range
of topics: school leaders’ strategies and practices for improving schools under
accountability demands and competition (Jabbar, 2015; Galdames et al., 2018; Ehren
& Shackleton, 2016); leadership and practices for school improvement in different
school contexts (Hallinger, 2018; Reed & Swaminathan, 2016); school leaders and
inclusive values for improving schools (Reed & Swaminathan, 2016), or a critical view
regarding low performing schools that change rapidly, known as “quick win” schools
(Meyers & Hitt, 2018).
The research on SI is linked not only to students, teachers, and leaders, but also to
their interactions and collaboration with diverse external agencies. Articles focused
on the relations between school leaders and policymakers (Daly et al., 2014; Dun-
away et al., 2014) and examined leaders’ responses and challenges in facing diverse
policies (Frankenberg, 2015; Gaertner et al., 2014; Liou, 2016) or professionals from
out of school (Swaffield, 2015). Other articles focused on promoting effective part-
nership between schools for school improvement as between high and low achieving
schools (Muijs, 2015), collaboration between schools in a divided society in Ireland
(Duffy & Gallagher, 2015), and collaboration between schools, which is a challenge
under policies that favor accountability and competition between schools (Ehren &
Perryman, 2018; Frankenberg et al., 2015). Armstrong and Ainscow (2018), stressed
trust as an essential factor for between schools’ collaborations. An additional two
182 A. Addi-Raccah, T. Shahrabani and E. Segev

articles looked at the partnership between practitioners and researchers/academia

(Ainscow et al., 2016; Sheard & Sharples, 2016). Four articles were related to diverse
subjects that were less attached to the main narratives revealed in Figure 9.1, such as
measuring school climate (Bradshaw et al., 2014) or trust (Romero & Mitchell,
2018); addressing health issues (Rasberry et al., 2015) or analyzing SI policies in
Trinidad and Tobago by adopting a historical perspective (James, 2014). These
studies seem to reflect the current approaches that regard schools as not working
alone, but rather collaborating with diverse external agencies while developing a
consolidation within the school community. Since teachers, as noted above, were
found to be at the center of the narrative on SI, we further focus on their specific
sub-network in greater detail.
Figure 9.2 illustrates the main topics related to teachers. The top-right cluster,
labeled as “facilitating factors”, includes many words that represent the conditions for
teachers’ work. In this cluster, two research foci are mentioned: one is teachers’ work
relations with other teachers represented with words such as “trust”, “community”,
and “professional development”. In this case, teachers may hold leadership roles by
participating in the core functions of teaching and learning (York-Barr & Duke,
2004). Teachers thus, develop their expertise together with their peers, and generate
new ideas for the development of schools (Leithwood & Jantzi, 2000).

FIGURE 9.2 Semantic network of words related to “teachers”.

Note. The network resolution is 100 sentences and above. Network visualization and
calculations were performed with Visone (Brandes & Wagner, 2004).
 School improvement 183

The other focus is related to their relationship with the school leadership or the
management. It refers to school principals who enhance teachers’ role in schools by
means of transformational or distributive leadership and capacity to manage change,
as found in prior studies (Leithwood & Jantzi, 2006). The relations between the
different enabling factors were found to be weak, indicating that studies often
choose only one of these foci, and do not offer a comprehensive analysis of the
different elements that support teachers’ work.
On the top-left, labeled “teachers’ practices”, is a cluster of general school
practices and processes of SI, including words such as “learning”, “teaching”,
“support”, and “development”. This cluster reflects teachers’ practices that are
related to students’ learning processes. At the bottom of the network, the cluster
labeled “students’ outcomes” focuses on cognitive outcomes such as “achieve-
ment”, “scores”, “test”, and “assessment”. Here too, the relations between the
terms in this cluster are relatively weak, indicating that studies employed different
factors to examine students’ outcomes. Apparently, the narrative emerging from
Figure 9.2 reflects the view that teachers are responsible for students’ outcomes, as
confirmed by numerous studies (e.g., Kyriakides & Creemers, 2009). Further, as
already indicated, the relationship between school leaders and students is indirectly
mediated through the teachers. Here too, parents seem to be a marginal factor in
schools, associated mainly with students.

Discussion
In this chapter, we employed semantic network analysis to map the literature on SI
during the five years between 2014 and 2018. We found that the main narrative
on SI is a conservative one. The responsibility for school improvement lies in the
hands of the school staff: teachers, and school leaders, through the process of
development, learning, or teaching. The students are presented as more passive in
this process with emphasis on their academic achievements or test scores. Still, it
appears that schools do not operate as independent units but are rather presented as
part of a larger network that is composed of diverse agencies. Further, a relatively
weak relationship exists between leaders and students compared to teachers and
leaders or teachers and students. These findings corroborate with prior studies showing
that the impact of school leaders on students’ outcomes (mainly achievements) is
mediated through teachers (Tian & Huber, 2019) and their improvement’s practices.
More specifically, when looking at the sub-network of teachers, we identify three
dimensions as responsible for SI (see also Mincu, 2015).

Factors contributing to teachers’ work

Our semantic networks identified two groups of factors that contributed to tea-
chers’ work: One was their relationship and cooperation with other teachers, and
the other was their relationship with the management.
184 A. Addi-Raccah, T. Shahrabani and E. Segev

In terms of the relationship with the school principals, the literature on SI

stresses the importance of teachers’ training and competencies (Snoek & Volman,
2014) as well as school leaders’ training. Orphanos and Orr (2014), for instance,
show the importance of investment in leadership preparation, which improves their
practices and yields more positive teachers’ work conditions that are essential for
improving students’ learning. Other studies similarly focused on the role of the
school leader: Minckler (2014) stressed the importance of transformational leader-
ship (e.g., encouraging high performance, providing vision and inspiration, and
providing intellectual stimulation) in developing and establishing a physical and
cultural environment for SI. Liljenberg (2015) focused on the implications of dis-
tributive leadership in Swedish schools for organizational development and learning
as experienced by teachers’ leadership. The study revealed the tensions raised by
distributive leadership along with expanded collaboration and trust. School leaders
were also found to contribute to teachers’ work through the development of col-
lective innovation, as an additional measure of school climate (Buske, 2018) or
through enhancing collegial and collaborative relations in schools (e.g., through
programs in schools, Ford & Youngs, 2018). Finally, Sebastian et al. (2017) showed
that principal leadership’ pathway to improve school climate learning is mediated
by teacher leadership. Thus, principals may prioritize their efforts in improving
learning climate, and enable teachers to assist in these efforts by increasing their
leadership capacity toward climate-related processes.
In terms of the relationship between the teachers themselves, studies show that SI
is related to teachers’ involvement in setting their school culture and collaborative
climate. For example, Bragg & Manchester (2017) referred to the Creative Partner-
ships program for building school ethos from a grassroots approach by learning from
all members of the school community (e.g., teachers and students). This study
emphasized the process of sharing and integrating different perspectives for enabling
SI from within rather than from outside of school. The significance of internal or
bottom-up processes was in accordance with Gaertner et al. (2014), showing that in
Germany principals’ and teachers’ perceptions of school improvement was cultivated
by the school staff rather than external inspections.
Another aspect of teachers’ agency was related to teachers’ leadership. On this
issue, Snoek and Volman (2014) showed that teachers who were engaged in active
dialogues with their supervisors and teams about their roles in developing the school
could change the school culture and improve teaching and learning. While school
leaders were both colleagues and mentors (Holloway et al., 2018), it was found that
teachers engaged in more reflective dialogues and felt collectively responsible when
the leadership held a group-oriented approach (Vanblaere & Devos, 2018).

Teachers’ practices and learning processes

Another body of knowledge that emerged from the semantic network analysis
focused on teachers’ diverse practices that were assumed to be related to students’
outcomes. Following the “third wave” of the school improvement initiative,
 School improvement 185

which emphasized teaching and learning standards, Hadfield and Jopling (2016)
refer to professional learning processes for improving students’ achievements. Other
studies addressed the importance of professional learning communities and collabora-
tion between teachers as a core component that contributes to school improvement,
enables pedagogical discussions, avoids hierarchical relations (Jäppinen et al., 2016),
and decreases bullying (Kyriakides et al., 2014). Even though Watson (2014) criticized
the role of professional learning communities, Voelkel and Chrispeels (2017) found
that it helped teachers’ collective efficacy and further improved students’ achieve-
ments. Similarly, Mitchell and Sackney (2016) examined teachers’ practices in a “living
system”, which were found to support authentic teaching and learning for students’
wellbeing. Studies also emphasized practices such as data use to understand students’
needs within the context of accountability demands (Beaver & Weinbaum, 2015; van
Geel et al., 2016; Schildkamp et al., 2017). This goes along with Lancer (2015), who
stressed the importance of basing professional development and learning on design
principles and theory in contrast to relying on conventional wisdom.

Assessments and students’ outcomes

The underlying assumption of most studies on SI is that improvement could be
quantitatively measured through regular assessments and tests of the students. Our
analysis showed that the theme of students’ tests, assessments, and scores was very
dominant in the narrative. Looking in-depth at the articles, however, we did not
find many articles that actually measured students’ outcomes and also referred to
teachers (data used in van Geel et al., 2016 and in Kyriakides et al., 2014).

Conclusions and future directions

Based on the semantic network analysis of recent research and the in-depth illus-
tration of a sample of articles, some conclusions could be drawn. First, within the
research on SI, teachers hold the most prominent role. As shown above, studies
examined various dimensions of teachers’ work, with an emphasis on their exten-
ded role as team leaders and colleagues. In this regard, we exemplify through the
content analysis of the articles the efforts of the educational system to improve
teachers’ work through professional development and establishing professional
learning communities that characterize 21st-century schools. Hence SI literature
mainly focuses on the perspective of schools as a setting for collective actions and
centers on system-level efforts for improvement (Hargreaves, 2012; Cheng, 2011;
Hopkins et al., 2014). Second, emphasis was given to the internal processes within
the schools, pointing to the relationships between school principals and teachers.
Yet, new directions seem to be emerging in the SI literature with the growing
realization of school networks and collaboration with external agencies for school
improvement (Hargreaves, 2012), suggesting heterarchical organizational patterns
in schools. While research on SI addresses both external and internal practices,
there is still the need for a comprehensive understanding of the relationships between
186 A. Addi-Raccah, T. Shahrabani and E. Segev

the diverse practices. Further, in the studies included in our sample, only a few stu-
dies actually measured students’ educational outcomes, and when doing so they
mainly focused on achievements. More importantly, there is hardly any emphasis on
life skills, social capital, or social and universal values. While achievements are sig-
nificant, there is also a need to address the students’ wellbeing, social, and emotional
outcomes, which are essential for their future success (Alzahrani et al., 2019;
Kanopka et al., 2020). Apparently, new directions of research on SI may be needed
in order to encompass the changes occurring in schools and in teachers’ work. Future
studies should take into account more broadly the relations between teachers and
school principals, expand the investigation of external actors, including parents, and
address diverse cognitive and non-cognitive students’ outcomes.

Tips and lessons for the use of semantic network analysis

1. Semantic network analysis that is based on a large body of articles

yields a very large volume of words. As the decision on which words
are to be included in the analysis is crucial, experts in the field have to
be active in this stage. Except for their frequency and conceptual cri-
teria for the word inclusion, it is advised that at least two experts in the
topic being examined will be involved in this process. Each expert will
evaluate the words to be included in the analysis and then the two
evaluations can be compared. For reaching reliability in the final list of
words to be included in the network analysis, it is suggested to exam-
ine the context of the chosen words within the text.
2. It is important to define the central dimensions of semantic networks. In
our case, we address three dimensions: actors, school processes, and
school outcomes. These dimensions were based on a simple and general
model related to school improvement. Having such a model in mind helps
to disclose the dominant focal points related to the topic of school
improvement, while leaving room for diverse aspects and expressions to
be included under these dimensions. In this way we could, for example,
identify the central actors related to school improvement (teachers versus
school principals), further explore the dynamic between them, and in
addition, point on gaps in the literature (e.g., in terms of school outcomes)
along with emerging factors related with school improvement.
3. As our semantic network analysis was based on articles, a meaningful
interpretation of the networks was achieved by a complementary qualita-
tive content analysis of a sample of articles. We chose a random sample of
articles to exemplify and provide an in-depth insight into the words and
relations between words emerging from the networks. A complementary
content analysis enabled to contextualize the topics that emerged in the
networks and offered an added value to the entire study.
 School improvement 187

References
Addi-Raccah, A., Amar, J., & Ashwal, Y. (2018). Schools’ influence on their environment: The
parents’ perspective. Educational Management Administration & Leadership, 46(5), 782–799.
Ainscow, M., Dyson, A., Goldrick, S., & West, M. (2016). Using collaborative inquiry to
foster equity within school systems: opportunities and barriers. School Effectiveness and
School Improvement, 27(1), 7–23.
Alzahrani, M., Alharbi, M., & Alodwani, A. (2019). The effect of social-emotional competence
on children academic achievement and behavioral development. International Education
Studies, 12(12), 141–149.
Armstrong, P. W., & Ainscow, M. (2018). School-to-school support within a competitive edu-
cation system: views from the inside. School Effectiveness and School Improvement, 29(4), 614–633.
Azkiyah, S. N. (2017). Educational Effectiveness Research as the Knowledge Base of
Improving Education. Pertanika Journal of Social Sciences and Humanities, 25(3), 1019–1038.
Beaver, J. K., & Weinbaum, E. H. (2015). State test data and school improvement efforts.
Educational Policy, 29(3), 478–503.
Bellei, C., Vanni, X., Valenzuela, J. P., & Contreras, D. (2016). School improvement tra-
jectories: an empirical typology. School Effectiveness and School Improvement, 27(3), 275–292.
Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of com-
munities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 10, 10008.
Bradshaw, C. P., Waasdorp, T. E., Debnam, K. J., & Johnson, S. L. (2014). Measuring
school climate in high schools: A focus on safety, engagement, and the environment.
Journal of School Health, 84(9), 593–604.
Bragg, S., & Manchester, H. (2017). Considerate, convivial and capacious? Finding a language
to capture ethos in “creative” schools. Discourse: Studies in the Cultural Politics of Education,
38(6), 864–879.
Brandes, U. &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In
M. Jünger & P. Mutzel (eds.), Graph Drawing Software (pp. 321–340). New York:
Springer-Verlag.
Buske, R. (2018). The principal as a key actor in promoting teachers’ innovativeness–analyzing
the innovativeness of teaching staff with variance-based partial least square modeling. School
Effectiveness and School Improvement, 29(2), 262–284.
Cheng, Y. C. (2011). Towards the 3rd wave school leadership. Revista de Investigación Educativa,
29 (2), 252–275.
Chernyshenko, O., M. Kankaraš and F. Drasgow (2018), “Social and emotional skills for
student success and wellbeing: Conceptual framework for the OECD study on social and
emotional skills”, OECD Education Working Papers, No. 173, Paris: OECD Publishing.
doi:10.1787/db1d8e59-en.
Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J., Mood, A. M., Weinfeld, F.,
& York, R. L. (1966). Equality of educational opportunity (Report No. OE-3800).
Washington, DC: National Center for Educational Statistics.
Creemers, B., Scheerens, J., & Reynolds, D. (2000). Theory development in school effective-
ness research. In The International Handbook of School Effectiveness Research (pp. 283–298).
London: Routledge.
Daly, A. J., Finnigan, K. S., Jordan, S., Moolenaar, N. M., & Che, J. (2014). Misalignment
and perverse incentives: Examining the politics of district leaders as brokers in the use of
research evidence. Educational Policy, 28(2), 145–174.
Danowski, J. A. (1993). Network analysis of message content. In W. D. Richards & G. A.
Barnett (eds.), Progress in Communication Sciences (Vol. 12, pp. 198–221). Norwood, NJ: Ablex
Publishing.
188 A. Addi-Raccah, T. Shahrabani and E. Segev

Duffy, G., & Gallagher, T. (2015). Collaborative evolution: The context surrounding the
formation and the effectiveness of a school partnership in a divided community in
Northern Ireland. Research Papers in Education, 30(1), 1–24.
Dunaway, D. M., Bird, J. J., Wang, C., & Hancock, D. (2014). Superintendents’ perceptions
of the school improvement planning process in the southeastern USA. Educational Man-
agement Administration & Leadership, 42(4), 460–473.
Ehren, M. C., & Shackleton, N. (2016). Mechanisms of change in Dutch inspected schools:
comparing schools in different inspection treatments. British Journal of Educational Studies,
64(2), 185–213.
Ehren, M., & Perryman, J. (2018). Accountability of school networks: Who is accountable to
whom and for what? Educational Management Administration & Leadership, 46(6), 942–959.
Ford, T. G., & Youngs, P. A. (2018). Creating organizational structures to facilitate collegial
interaction among teachers: Evidence from a high-performing urban Midwestern US
district. Educational Management Administration & Leadership, 46(3), 424–440.
Frankenberg, E., McDermott, K. A., DeBray, E., & Blankenship, A. E. (2015). The new
politics of diversity: Lessons from a federal technical assistance grant. American Educational
Research Journal, 52(3), 440–474.
Fullan, M. (2009). Large-scale reform comes of age. Journal of Educational Change, 10(2–3),101–113.
Gaertner, H., Wurster, S., & Pant, H. A. (2014). The effect of school inspections on school
improvement. School Effectiveness and School Improvement, 25(4), 489–508.
Galdames, S., Montecinos, C., Campos, F., Ahumada, L., & Leiva, M. V. (2018). Novice
principals in Chile mobilizing change for the first time: Challenges and opportunities
associated with a school’s readiness for change. Educational Management Administration &
Leadership, 46(2), 318–338.
Gorard, S., Siddiqui, N., & See, B. H. (2015). An evaluation of the “Switch‐on Read-
ing”literacy catch‐up programme. British Educational Research Journal, 41(4), 596–612.
Gorard, S., Siddiqui, N., & See, B. H. (2017). What works and what fails? Evidence from
seven popular literacy “catch-up”schemes for the transition to secondary school in Eng-
land. Research Papers in Education, 32(5), 626–648.
Greatbatch, D., & Tate, S. (2019). School Improvement Systems in High Performing Countries.
Manchester: Department of Education.
Gumus, S., Bellibas, M. S., Esen, M., & Gumus, E. (2018). A systematic review of studies on
leadership models in educational research from 1980 to 2014. Educational Management
Administration & Leadership, 46(1), 25–48.
Hadfield, M., & Jopling, M. (2016). Problematizing lesson study and its impacts: Studying a
highly contextualised approach to professional learning. Teaching and Teacher Education, 60,
203–214.
Hallinger, P. (2018). Bringing context out of the shadows of leadership. Educational Man-
agement Administration & Leadership, 46(1), 5–24.
Hargreaves, A. (2012). Singapore: the Fourth Way in action ? Educational Research for Policy
and Practice, 11(1), 7–17.
Higham, R., & Booth, T. (2018). Reinterpreting the authority of heads: Making space for
values-led school improvement with the Index for Inclusion. Educational Management
Administration & Leadership, 46(1), 140–157.
Hill, P. T., & Guthrie, J. W. (1999). A new research paradigm for understanding (and
improving) twenty-first century schooling. Handbook of research on educational administration,
2, 511–524.
Holloway, J., Nielsen, A., & Saltmarsh, S. (2018). Prescribed distributed leadership in the era
of accountability: The experiences of mentor teachers. Educational Management Adminis-
tration & Leadership, 46(4), 538–555.
 School improvement 189

Hopkins, D., Ainscow, M., & West, M. (1994). School Improvement in An Era of Change
(Chapter 9). Great Britain: Redwood Books.
Hopkins, D., Stringfield, S., Harris, A., Stoll, L., & Mackay, T. (2014). School and system
improvement: A narrative state-of-the-art review. School Effectiveness and School Improvement,
25(2), 257–281.
Jabbar, H. (2015). “Every kid is money”: market-like competition and school leader strategies
in New Orleans. Educational Evaluation and Policy Analysis, 37(4), 638–659.
James, F. (2014). The school improvement policy context in Trinidad and Tobago. School
Effectiveness and School Improvement, 25(3), 469–485.
Jäppinen, A. K., Leclerc, M., & Tubin, D. (2016). Collaborativeness as the core of profes-
sional learning communities beyond culture and context: evidence from Canada, Finland,
and Israel. School Effectiveness and School Improvement, 27(3), 315–332.
Kanopka, K., Claro, S., Loeb, S., West, M. R., & Fricke, H. (2020). What Do Changes in
Social-Emotional Learning Tell Us About Changes in Academic and Behavioral Outcomes? Policy
Analysis for California Education. Retrieved on 1 October 2020 from https://www.edpoli
cyinca.org/sites/default/files/2020-07/pace_policy_brief_-_change_sel_202007_1.pdf.
Keddie, A. (2014). School collaborations within the contemporary English education system:
possibilities and constraints. Cambridge Journal of Education, 44(2), 229–244.
Kelcey, B., & Shen, Z. (2016). Multilevel design of school effectiveness studies in sub-
Saharan Africa. School Effectiveness and School Improvement, 27(4), 492–510.
Kovačević, J. and Hallinger, P. (2019). Leading school change and improvement: A bibliometric
analysis of the knowledge base (1960–2017). Journal of Educational Administration, 57(6),
635–657.
Kyriakides, L., & Creemers, B. P. (2009). The effects of teacher factors on different out-
comes: Two studies testing the validity of the dynamic model. Effective Education, 1(1),
61–85.
Kyriakides, L., Creemers, B. P., Muijs, D., Rekers-Mombarg, L., Papastylianou, D., Van
Petegem, P., & Pearson, D. (2014). Using the dynamic model of educational effectiveness
to design strategies and actions to face bullying. School effectiveness and school improvement,
25(1), 83–104.
Lancer, J. R. (2015). The meaning of quality professional learning for school improvement:
articulating a coherent vision rooted in a theoretical perspective on learning. School Effec-
tiveness and School Improvement, 26(4), 639–667.
Leithwood, K. & Jantzi, D. (2000). Principal and teacher leadership effects: A replication.
School Leadership & Management, 20(4), 415–434.
Leithwood, K., & Jantzi, D. (2006). Transformational school leadership for large-scale
reform: Effects on students, teachers, and their classroom practices. School effectiveness and
school improvement, 17(2), 201–227.
Leithwood, K., Jantzi, D., & Steinbach, R. (1999). Changing Leadership for Changing Times.
London: McGraw-Hill Education.
Liljenberg, M. (2015). Distributing leadership to establish developing and learning school
organisations in the Swedish context. Educational Management Administration & Leadership,
43(1), 152–170.
Liou, Y. H. (2016). Tied to the common core: Exploring the characteristics of reform advice
relationships of educational leaders. Educational Administration Quarterly, 52(5), 793–840.
Louis, K.S., Toole, J., & Hargreaves, A. (1999). Rethinking school improvement. In J.
Murphy & K.S. Louis (Eds.), Handbook of Research on Educational Administration (2nd ed.,
pp. 251–275). San Francisco: Jossey Bass.
May, H., Huff, J., & Goldring, E. (2012). A longitudinal study of principals’ activities and
student performance. School Effectiveness and School Improvement, 23(4), 417–439.
190 A. Addi-Raccah, T. Shahrabani and E. Segev

McInerney, C., Bird, N., & Nucci, M. (2004). The flow of scientific knowledge from lab to
the lay public: The case of genetically modified food. Science Communication, 26(1), 44–74.
Meyers, C. V. & Hitt, D. H. (2018). Planning for school turnaround in the United States: an
analysis of the quality of principal-developed quick wins. School Effectiveness and School
Improvement, 29(3), 362–382.
Minckler, C. H. (2014). School leadership that builds teacher social capital. Educational
Management Administration & Leadership, 42(5), 657–679.
Mincu, M. E. (2015). Teacher quality and school improvement: what is the role of research?
Oxford Review of Education, 41(2), 253–269.
Mitchell, C., & Sackney, L. (2016). School improvement in high-capacity schools: Educa-
tional leadership and living-systems ontology. Educational Management Administration &
Leadership, 44(5), 853–868.
Muijs, D. (2010). A fourth phase of school improvement? Introduction to the special issue
on networking and collaboration for school improvement. School Effectiveness and School
Improvement, 21(1), 1–3.
Muijs, D. (2015). Improving schools through collaboration: a mixed methods study of
school-to-school partnerships in the primary sector. Oxford Review of Education, 41(5),
563–586.
Ni, Y., Yan, R. & Pounder, D. (2017). Collective Leadership: Principals’ Decision Influence
and the Supportive or Inhibiting Decision Influence of Other Stakeholders. Educational
Administration Quarterly, 54(2), 216–248.
Nurkolis, N., & Sulisworo, D. (2018). School Effectiveness Policy in the Context of Edu-
cation Decentralization. Journal of Education and Learning, 12(2), 244–252.
OECD. (2019). Education Policy Outlook 2019: Working Together to Help Students Achieve their
Potential. Paris: OECD Publishing.
Orphanos, S., & Orr, M. T. (2014). Learning leadership matters: The influence of innova-
tive school leadership preparation on teachers’ experiences and outcomes. Educational
Management Administration & Leadership, 42(5), 680–700.
Pashiardis, P., & Brauckmann, S. (2019). New public management in education: A call for
the edupreneurial leader? Leadership and policy in schools, 18(3), 485–499.
Peurach, D. J., Glazer, J. L., & Winchell Lenhoff, S. (2016). The developmental evaluation
of school improvement networks. Educational Policy, 30(4), 606–648.
Preston, C., Goldring, E., Guthrie, J. E., Ramsey, R., & Huff, J. (2017). Conceptualizing
essential components of effective high schools. Leadership and Policy in Schools, 16(4), 525–562.
Rasberry, C. N., Slade, S., Lohrmann, D. K., & Valois, R. F. (2015). Lessons learned from the
whole child and coordinated school health approaches. Journal of School Health, 85(11), 759–765.
Reed, L. C. & Swaminathan, R. (2016). An Urban School Leader’s Approach to School
Improvement: Toward Contextually Responsive Leadership. Urban Education, 51(9),
1096–1125.
Reynolds, D., Sammons, P., De Fraine, B., Van Damme, J., Townsend, T., Teddlie, C., &
Stringfield, S. (2014). Educational effectiveness research (EER): a state-of-the-art review.
School Effectiveness and School Improvement, 25(2), 197–230.
Romero, L. S., & Mitchell, D. E. (2018). Toward understanding trust: A response to Adams
and Miskell. Educational Administration Quarterly, 54(1), 152–170.
Schildkamp, K., Poortman, C., Luyten, H., & Ebbeler, J. (2017). Factors promoting and
hindering data-based decision making in schools. School effectiveness and school improvement,
28(2), 242–258.
Sebastian, J., Huang, H., & Allensworth, E. (2017). Examining integrated leadership systems
in high schools: Connecting principal and teacher leadership to organizational processes
and student outcomes. School Effectiveness and School Improvement, 28(3), 463–488.
 School improvement 191

Segev, E. (2020). Semantic network analysis: Detecting prevailing themes and biases in
international news and social media. Sociology Compass, 14(4), e12779.
Sheard, M. K., & Sharples, J. (2016). School leaders’ engagement with the concept of evidence-
based practice as a management tool for school improvement. Educational Management
Administration & Leadership, 44(4), 668–687.
Snoek, M., & Volman, M. (2014). The impact of the organizational transfer climate on the
use of teacher leadership competences developed in a post-initial Master’s program.
Teaching and Teacher Education, 37, 91–100.
Spillane, J. P. (2012). Distributed leadership (Vol. 4). London: John Wiley & Sons.
Stoll, L., & Fink, D. (1994). School effectiveness and school improvement: Voices from the
field. School Effectiveness and School Improvement, 5(2), 149–177.
Swaffield, S. (2015). Support and challenge for school leaders: Headteachers’ perceptions of
school improvement partners. Educational Management Administration & Leadership, 43(1),
61–76.
Tian, M., & Huber, S. G. (2019). Mapping educational leadership, administration and
management research 2007–2016. Journal of Educational Administration, 58(2), 129–150.
Valenzuela, J. P., Bellei, C., & Allende, C. (2016). Measuring systematic long-term trajec-
tories of school effectiveness improvement. School Effectiveness and School Improvement, 27
(4), 473–491.
Van Geel, M., Keuning, T., Visscher, A. J., & Fox, J. P. (2016). Assessing the effects of a
school-wide data-based decision-making intervention on student achievement growth in
primary schools. American Educational Research Journal, 53(2), 360–394.
Vanblaere, B., & Devos, G. (2018). The role of departmental leadership for professional
learning communities. Educational administration quarterly, 54(1), 85–114.
VoelkelJr, R. H., & Chrispeels, J. H. (2017). Understanding the link between professional
learning communities and teacher collective efficacy. School Effectiveness and School
Improvement, 28(4), 505–526.
Watson, C. (2014). Effective professional learning communities? The possibilities for teachers
as agents of change in schools. British Educational Research Journal, 40(1), 18–29.
Wrigley, T. (2008). School Improvement in a neo-liberal world. Journal of Educational
Administration and History, 40(2), 129–148.
York-Barr, J., & Duke, K. (2004). What do we know about teacher leadership? Findings
from two decades of scholarship. Review of educational research, 74(3), 255–316.
10
IDENTIFYING PATTERNS IN
COMMUNICATION SCIENCE
Mapping knowledge structures using semantic
network analysis of keywords

Nico Pfiffner

Identifying patterns in communication science

In a fast-paced academic environment where publication cycles are becoming
shorter, and the number of publication outlets is increasing, assessing the state
of a research domain is an important task. Not only can such an assessment
highlight general trends and longitudinal developments occurring in a research
area, but it also has the potential to unveil research gaps and blind spots where
there is a need for further studies (Tomasello et al., 2010). In this regard,
communication science is an interesting research area to investigate. First, it is
often characterized as interdisciplinary and fragmented, which is also reflected
in ongoing discussions about what exactly constitutes the field of commu-
nication science. Second, the field is characterized by a rapid growth and
change (Simonson, 2015), which makes it even more important to regularly
monitor the state of the domain. On the one hand, these observations can be
utilized to uncover areas of strength and weakness, or abundance and scarcity
in research, thereby having implications on future research directions (Kam-
hawi & Weaver, 2003). On the other hand, it can also show the core areas of
a thematically dispersed field and reveal how different sub-fields are (or are
not) connected.
Consequently, this study explores the knowledge structure underlying com-
munication science by constructing and analyzing a semantic network based on
keywords assigned to peer-reviewed journal articles that have been published in
leading communication journals with different thematic foci. Ultimately, the
goal is to answer the following question: What are the structural commonalities
and most important topics across different sub-domains of communication
science?

DOI: 10.4324/9781003120100-11
 Patterns in communication science 193

The history of communication science

To gain a better understanding of communication science as a research domain, it
is important to take the history of the field into consideration. Communication
science is a comparatively young institutionalized academic research domain.
Although communication has always been the building block of our society, until
the 1940s it was investigated across a variety of different disciplines that had little
connections among each other (Bryant & Pribanic-Smith, 2010). During this early
period, five important but largely distinct fields of communication research could
be identified: Communication and political issues, communication in social life,
social-psychological analyses of communication, communication and education,
and lastly commercial communication research (Delia, 1989).
After World War II, these research strands have gradually consolidated, driven by
the growing interest in communication research, and the stronger interpersonal con-
tacts among leading communication scholars (Bryant & Pribanic-Smith, 2010; Delia,
1989). The first departments and institutes of communication were founded at major
universities in the US at the beginning of the 1950s (Wartella, 1996). During the same
time, the predecessor of the International Communication Association (ICA) was
established, bringing together communication scholars with different thematical
foci and backgrounds, greatly contributing to the formation of a comprehensive
communication science domain (McCroskey & Richmond, 1996). Although the
organizational establishment of communication science as a scientific domain started
later in most other parts of the world (Koivisto et al., 2010), over the following years
communication science witnessed a rapid growth and today is institutionalized at
many universities around the world, and regional communication associations have
been formed (Simonson, 2015).
Still, the field of communication science nowadays is often regarded as fragmented
and “hyper-specialized”, making it hard to find a common definition of commu-
nication science or even a shared terminology (Waisbord, 2019). There seems to be no
generally accepted understanding of what precisely communication sciences are and
how the boundaries and internal divisions of the field are defined (Craig, 1993;
Donsbach, 2006; Jiménez & Guillem, 2012; Reardon & Rogers, 1988).
Against the background of its historical development, this fragmentation does not
come as a surprise. Communication studies have been influenced by a variety of
other disciplines, such as sociology, psychology, and political science, but also
domains like mathematics or engineering (Berger et al., 2010; Craig, 1999). Theo-
retical frameworks and methods from these fields have frequently been adapted and
further developed by communication scholars (Berger et al., 2010). While there
seems to be an agreement that the original foundation of communication science lies
on the study of speech and the study of mass media (Khan et al., 2016), its scope was
later expanded to additional core topics such as human communication, advertising,
public relations, and new media (Chung et al., 2013). Consequently, communication
studies have developed in many different directions, while witnessing a steady
growth in the number of publications (Barnett et al., 2011; Littlejohn & Foss, 2009).
194 N. Pfiffner

As a result, communication science is a highly diverse field characterized by multiple

traditions and research interests (Waisbord, 2019), which makes it difficult to char-
acterize it as a whole. This is exactly why it is crucial to monitor the current state of
communication science and identify its core elements.

Determining the knowledge structure of a scientific domain

Since the advent of electronic publication and the digitalization of information
archives, mapping the knowledge structures of scientific domains has gained great
attention from researchers across disciplines (Evans & Foster, 2011). This development
is also driven by the fact that digitalization has led to an increase in research publica-
tions and a more complex scientific environment which fosters the interest in the
exploration of patterns in scientific knowledge structures (Evans & Foster, 2011).
In any scientific field, new insights are the result of a continuous social and intel-
lectual process of reconstructing and recombining pre-existing fragments of knowl-
edge, which is to a great extent mediated by scientific publications. These publications
themselves build on existing concepts captured in previously published works and
usually undergo a peer review validation process (Leydesdorff & Milojević, 2015).
Peer-reviewed research documents can be regarded as reliable carriers of knowledge,
and thus observing their evolution also means observing the formation and evolution
of scientific knowledge. By studying scientific publications, this evolutionary process
can be retraced to reveal the knowledge structure underlying a scientific domain
(Assefa & Rorissa, 2013). Developed in the late 1970s, Scientometrics constitutes a
whole research area devoted to the quantitative study of science that aims, among
other things, to map and understand the intellectual structures of scientific domains
(Gupta & Bhattacharya, 2004; Kim & Zhu, 2018). Studies with this goal usually utilize
subject terms, information about authors, or scientific publications (i.e., journals or
single articles) to discover the underlying structures of a field (Leydesdorff &
Milojević, 2015). Generally, there are two common types of scientometric
approaches to mapping a scientific field: (1) Examining citation structures or (2)
investigating commonalities of the content of scientific publications. Both approaches
are based on the co-occurrence of two entities of interest, or in other words, on
networks (Leydesdorff & Milojević, 2015).
Citation structures are often investigated through journal-to-journal citation maps or
co-citation maps. While the former is suited to uncover the macro-level structure of
a field by revealing the linkages among journals covering different sub-fields, the
latter provides a more detailed picture of the research front by clustering documents
that cite each other or are cited in the same context together (Boyack & Klavans,
2010; Gupta & Bhattacharya, 2004).
Content commonalities are usually investigated by identifying pairs of words, topics,
or concepts that occur together. This approach assumes that a link exists between
two distinct concepts if they co-occur in the same document. In this way, looking
at a large corpus of documents, topic clusters can be uncovered that represent the
sub-fields of a research domain (Gupta & Bhattacharya, 2004).
 Patterns in communication science 195

One example of an approach building on content commonalities that is parti-

cularly relevant for the study at hand are keyword networks. Keywords are
assigned to publications by authors or editors to highlight the most relevant topics
covered in its main text. Therefore, it can be assumed that the keywords of a
publication are an adequate representation of its content and that two keywords
co-occurring in the same paper indicate a link between the two underlying con-
cepts (Ding et al., 2001). Constructing a keyword network that includes many
publications belonging to a certain scientific field can yield a knowledge map
revealing the core topics and main areas of the field.
In summary, the main goal of all these mapping approaches is to capture and
analyze the structures of scientific domains to better understand (a) what the main
research topics of a particular domain are, (b) how these different topics are related to
each other, and (c) how a scientific field has developed over time (Waltman et al.,
2010). This contributes to understanding the dynamics of science in general and can
inform decisions on the allocation of financial and attention resources (Leydesdorff &
Milojević, 2015).

The structure of communication science

In the past, the field of communication science has often been described as a
dichotomy of two main strands: The study of speech and interpersonal commu-
nication, and the study of mass media communication (McMahan, 2004; Reardon &
Rogers, 1988; Rogers, 1999). But the technological advancements over the last
decades and the growing use of social media have blurred the lines between inter-
personal and mass communication. Therefore, this dichotomization has become too
simplistic and does not adequately reflect the current state of the field (O’Sullivan &
Carr, 2018). To examine the structure of communication science further, different
strategies can be applied: First, the structure can be derived top-down from organi-
zational divisions of communication departments and associations. Second, building
on scientometric approaches, the core topics of the field can be determined bottom-
up by investigating the contents of communication science publications. Third, co-
citation networks of communication publications can be investigated to reveal more
about the interconnectedness of different sub-domains.

Thematic divisions
A starting point to identify the core areas of communication science are the thematic
divisions of communication associations. Currently, both the ICA and the European
Communication Research and Education Association (ECREA) list 24 (ECREA,
2020; ICA, 2020) and the Asociación de Investigatores de la Comunicaión (ALAIC)
19 different thematic divisions (ALAIC, 2020). An overview of the thematic topics
across these three associations already shows the diversity of communication science
as the associations diverge notably in the definition of their core divisions. A study
examining joint division memberships in the ICA found three underlying dimensions
196 N. Pfiffner

differentiating the ICA’s divisions: The first one separates mass media and inter-
personal communication, the second differentiates the humanities from the social
sciences, and the third showed a separation between theoretical and applied research
(Barnett & Danowski, 1992).
A more recent investigation of the structure of the division co-memberships of
scholars in the ICA and a Korean communication association (KSJCS) showed that
in both cases mass media and new media are the core thematical divisions along with
divisions focusing on political and international or intercultural issues. Furthermore, a
dichotomy between interpersonal and mass media communication can still be
observed in the membership structures of both associations (Chung et al., 2009).

Content characteristics
A second approach is to examine the content of communication publications. Studies
in this vein often attempt to classify communication studies based on their different
methods, theories, types of media under investigation, and their various themes.
Applied methods. A meta-analysis of articles published between 1980 and 1999 in
major mass communication journals found that (mass) communication science is
dominated by quantitative studies, with survey and content analyses being the most
prominent methods (Kamhawi & Weaver, 2003). The same study estimated that
qualitative methods were applied in approximately 30% of the published articles
(Kamhawi & Weaver, 2003), while another study covering the same period esti-
mated the ratio of quantitative to qualitative studies at about 60:40 (Trumbo,
2004). More recent analyses estimate the ratio to be closer to 50:50 (Borah, 2017;
Lee et al., 2016) or even in favor of qualitative studies, although this depends on
the sub-area under investigation. In summary, qualitative methods seem to have
gained popularity over the last years compared to the end of the 20th century.
Theories in communication science. Many authors have found that often there are no
explicit theories referenced in communication publications (Kamhawi & Weaver,
2003; Potter & Riddle, 2007). Nevertheless, based on an examination of four major
communication journals, Chung et al. (2013) identified five broader groups of
communication theories: Mass communication theories (e.g., framing, agenda-setting, uses
and gratifications, or cultivation theory), theories related to communication technologies (e.g.,
diffusion of innovation), interpersonal communication theories (e.g., social presence theory,
or uncertainty reduction theory), persuasion theories (e.g., social cognitive theory, theory
of planned behavior, or the elaboration likelihood model), and organizational commu-
nication theories (e.g., system theory).
In a study investigating mentions of communication theories in Google web
search results, Khan et al. (2016) identified nine theoretical sub-strands of com-
munication science: Interpersonal and relationship theories, mass media theories, commu-
nication process theories, language and linguistics theories, public relations, advertising,
marketing and consumer behavior theories, organizational communication theories, commu-
nication and information technology theories, health communication theories, and media,
culture, and society theories (Khan et al., 2016). Overall, framing seems to be the most
 Patterns in communication science 197

prominent and most frequently referenced theory in communication science

(Chung et al., 2013; Khan et al., 2016).
Types of media. As communication technologies have changed over the last
decades, so have the most frequently investigated media types (Potter & Riddle,
2007). This includes a slight decrease in traditional media research (mainly on
television and newspapers) and a marginal increase in internet research from
1980 to 1999 (Kamhawi & Weaver, 2003). After this period, however, elec-
tronic media and the internet have become increasingly popular and are now
the most researched forms of communication (Günther & Domahidi, 2017;
Potter & Riddle, 2007).
Main thematical topics. Topic modeling is often used to infer the core topics of
communication research from the text components of scientific publications.
According to an analysis that applied this quantitative approach to the abstracts of
articles published in 19 major communication journals from the 1930s to 2014, the
15 most frequent topics in the field of communication science were education,
marketing and PR, media use, comparative research and media stereotypes, health, new
media, survey and interview research, media violence and media effects, religion, family and
development, youth at risk, trust, privacy, and credibility, relationships, and crisis and conflict
(Günther & Domahidi, 2017). In another study that applied a keyword network
analysis based on communication articles published between 1980 and 2013, 16
thematic areas were identified: Children, psychological aspects, news, audiences, survey,
advertising, health, (interpersonal) relationships, gender, discourse, telephone communication,
public relations, telecommunications, public opinion, activism, and internet research (Mon-
tero-Díaz et al., 2018).
When comparing the results of the two mentioned studies, there are some topics
that are consistent across both and others that are only mentioned in one of the
studies. In both cases, marketing/advertising and public relations highlight the eco-
nomical perspective, relationships emphasize the interpersonal aspect, and health
communication is identified as a distinguishable subject. Survey research is also a
dominant methodological approach identified by both studies. Furthermore, both
studies find topics related to children and adolescents (family and development, youth
at risk, and children), recipients of mass communication (media use and audience), psy-
chological media effects (media violence, media effects, and psychological aspects), as
well as technology-related studies (new media, telephone communication, tele-
communication, and internet). On the other hand, more specific topics such as
religion, trust, privacy and credibility, and crisis and conflict, or gender, activism, and public
opinion are only mentioned in one of the studies. This indicates that there are some
core aspects of communication science that are relatively stable and others that are
less established thematic areas.
Furthermore, some of the core topics identified in the two studies can also be
found in some of the communication associations’ divisions, but many of them do
not have a direct equivalent. This could mean that there are certain meta-topics
that span across different areas forming a topical layer that connects seemingly dis-
tinct sub-fields of communication science.
198 N. Pfiffner

Co-citation patterns
Investigating co-citation patterns is often done through network analysis and
usually examines citation patterns between disciplines, journals, or authors.
Earlier studies looking at citation structures among communication journals did
not identify general communication journals that were frequently cited across
sub-disciplines, thus bridging different communication areas. Instead, it was
found that communication journals were rather leaning toward one area of
specialization, leaving the field in a relatively fragmented state (So, 1988). In
these studies, psychology was often identified as the most influential discipline
that was most often cited by communication publications (Park & Leydesdorff,
2009; So, 1988). However, a more comprehensive study found that journals
exclusively listed as communication journals tend to be more central in the citation
network than interdisciplinary journals that are also connected to different other
disciplines (Barnett et al., 2011). This could indicate that there is a core or
common understanding of communication science that different sub-disciplines rely
on. Furthermore, the main non-communication journals referenced by commu-
nication studies can be divided into three clusters: psychology, socio-politics, and
gender studies.

Research questions
Naturally, the previous overview does not cover all the attempts to investigate
and map communication science. However, synthesizing the presented results,
communication science still seems to span across many sub-fields that are
characterized by their own thematic foci. At the same time, some aspects of
communication, such as the uses and implications of digital communication
seem to be relevant across many sub-fields, possibly building the bridging
backbone of the domain. Still, there is no clear answer to how the field is
currently structured, how much sub-domains overlap, and what their common
denominators are. Therefore, this study seeks to address the following research
questions:

RQ1: What are the structural commonalities among different sub-domains of

communication research?
RQ2: What are the most prominent topics covered by communication research in
general and in each of its sub-domains in particular?

Answering these questions will contribute to a better understanding of the structure

and the identity of communication science. This can help to address research gaps
and to reveal new combinations of thus far distinct topics by providing a compre-
hensible overview of the research area and its sub-dimensions. A semantic network
analysis based on the keywords of peer-reviewed articles is applied to investigate
these questions.
 Patterns in communication science 199

Method

Keyword network analysis

Semantic network analysis was applied to analyze keywords of academic papers in
leading communication journals. For this study, peer-reviewed journal articles were
preferred over other data sources. First, journal articles are regarded as the main
channel for reporting current research (Kamhawi & Weaver, 2003), and suitable to
reflect current trends and the longitudinal evolution of communication research
(Riffe & Freitag, 1997; Shaw et al., 2000). Second, since these articles undergo a
peer review process, it can be assumed that they satisfy scientific quality standards,
and that the reported viewpoints are supported by more than one individual rather
than depicting the opinion of a single person.
Keywords from academic papers can either be (manually or algorithmically)
extracted from the full texts, abstracts, or titles, or they can be assigned by the
authors. Keywords that are deliberately selected by the author or the journal editor
already contain an attribution of importance. They are a result of an interpretation
and classification process executed by domain experts, a step that in other research
settings is often covered by a computational content analysis based on statistics
derived from the linguistic corpus. A possible advantage of a human expert inter-
pretation over a computational interpretation is that the former also includes more
subtle factors such as the interests of the author (Yang & Jung, 2016) or expectations
of which keywords might seem attractive to the research community at the time of
publication, thereby capturing the attitude of the expert and the community as well
as the current Zeitgeist.
All things considered, although keywords might be one of the smallest units of
analysis compared to other applications of semantic network analysis, they already
contain an importance attribution which usually is not the case for single words
or concepts extracted from a larger text. Therefore, this study will be based on
expert-assigned keywords attributed to scientific research articles published in
peer-reviewed communication journals.

Sample
Based on the research questions, as well as the considerations stated above, the
following selection criteria were defined: The sample must include (a) peer-
reviewed communication journals with (b) different thematic foci, and these jour-
nals must (c) provide author or editor keywords for their published articles.
Lists of top communication journals according to the h5-index as reported by
Google Scholar metrics and the InCites Journal Citation Report by Clarivate were
used as a starting point. Next, journals that did not satisfy one or more of the defined
sample criteria consequently excluded. In the end, a total of eight communication
journals were included in the sample, with one journal being a general commu-
nication journal (Communication Research) and the other seven each covering a specific
200 N. Pfiffner

sub-domain of communication science. Although some of the most influential

communication journals could not be included in the analysis because they did not
comply with all selection criteria mentioned above, the current sample includes
communication journals covering a wide range of sub-domains which makes it ideal
to investigate the knowledge structure that connects different sub-fields. The names
as well as the publishers and the respective sub-domain of the selected journals are
listed in Table 10.1.
All research articles published in these journals between 2008 and 2019 are
incorporated in this study.

Data collection and preparation

The data were collected using a script written in Python 3 that in a first step col-
lected all DOIs belonging to the journals in the sample as well as all available meta-
information associated with each DOI through the Crossref REST API (Crossref,
2020). Because the meta-data registered on Crossref did not include the keywords
belonging to an article, these data were in a second step directly scraped from the
respective journal websites. When retrieving information from the journal websites,
appropriate delay times were incorporated between requests and the request header
included the authors contact information to ensure a fair and respectful scraping
process. Subsequently, all documents that were not original research articles (e.g.,
book reviews, editorials, etc.) as well as articles that did not have any keywords
assigned to them were excluded. In total, 4731 articles remained in the sample and
were considered in the following analysis.
A total of 13,279 different keywords were initially assigned to these articles. To
make the analysis more comprehensible, a standardization of the keywords was
conducted as follows: Accented characters were converted to their “normal” form

TABLE 10.1 Communication journals in the sample

Journal name Abbr. Publisher Sub-domain # of articles

Communication Research CR SAGE General 592
New Media & Society NMS SAGE New Media 1235
Journalism Studies JS Taylor & Journalism 867
Francis
Political Communication PC Taylor & Politics 324
Francis
Journal of Media Psychology JMP Hogrefe Psychology 226
Public Relations Review PRR Elsevier Public Relations 871
Management Communication MCQ SAGE Management 271
Quarterly
Science Communication SC SAGE Science 345
Communication
4731
 Patterns in communication science 201

(e.g., “é” to “e”), punctuations contained in words were removed (e.g., “non-
binary” becomes “nonbinary”), keywords that contained multiple words separated
by a space were split into separate keywords (e.g., “quantitative methods” becomes
“quantitative” and “methods”), keywords existing in a defined list of stop-words
were excluded from the analysis,1 lemmatization was applied to reduce the inflec-
tional forms of a word to its root form (Ignatow & Mihalcea, 2017; e.g., “is” or
“am” become “be”, “transcripts” becomes “transcript”),2 and keywords categorized
as pronouns were excluded from the analysis.
Finally, the link list was created based on the clean list of keywords. A tie was
formed between two keywords if they appeared together in the same publication.
This link list was then used to construct an undirected network with the ties
weighted according to the number of articles a keyword pair appeared in the
sample. The complete unweighted keyword network, including all journals and
covering all articles published from 2008 to 2019, included 6040 nodes (keywords)
connected by 116,089 ties. Additionally, separate networks are constructed for each
journal.

Results
The result section is divided into two parts. First, a descriptive overview of the
structural unweighted network characteristics is presented and compared among
the different networks to identify structural similarities and differences (RQ1). The
second part focuses on the thematic structure of the networks, identifying the
actual keywords that are at the core of the aggregated communication science
network as well as at the core of the different sub-discipline networks (RQ2).

Structural network characteristics

When looking at the structural network characteristics, the density, the average
shortest path length, and the average clustering coefficient are the main measures to be
considered. The network’s density is computed by dividing the number of existing
edges by the number of all theoretically possible edges in the network. The closer
this measure is to 1, the more interconnected the keyword network is. The average
shortest path length describes the average length of the shortest paths between all
keywords. The information flow and exchange of ideas tends to be faster in net-
works with a shorter characteristic path length (Choi et al., 2011). For keyword
networks, a short average path length means that the concepts in a network
represented as nodes are closer to each other compared to a network with a longer
average path. The clustering coefficient describes the tendency of nodes in the net-
work to cluster together (Luke, 2015). When this coefficient is high, the network
consists of many tightly connected clusters, meaning that many keywords form
relatively independent sub-groups. But to avoid misinterpretation, the clustering
coefficient should usually be interpreted compared to the network density (Han-
neman & Riddle, 2011). In the case at hand, all these measures are computed based
202 N. Pfiffner

on the unweighted network. Table 10.2 provides an overview of these measures

for all the different networks in the analysis, as well as information on the number
of articles, nodes, and edges in each network.
Looking at the measures for the average shortest path length and the clustering
coefficient, no great differences or patterns can be observed among the networks.
However, regarding the network densities, there is a difference between the com-
plete aggregated network (labeled “Total” with a density of 0.006) and the indi-
vidual journal networks (in which the density ranges from 0.010 to 0.025).
Although all density scores are rather small, the densities of the journal networks
are all higher than the density of the complete aggregated network. This means
that the keywords in the journal networks are stronger connected among each
other compared to the ones in the aggregated network, which indicates that the
different journal networks indeed cover thematic sub-areas that are stronger con-
nected within than among each other. Another interesting difference is between
Political Communication (PC) with the highest density (density = 0.025) and New
Media & Society (NMS) with the lowest density (density = 0.010). This means that
the topics in the former journal are more homogenous than those of the latter.
Furthermore, the low densities together with the relatively high clustering coeffi-
cients indicate that the networks consist of many closely related sub-clusters that are at
the same time relatively sparsely connected to each other (Choi et al., 2011). This is
most likely because each included article adds a tightly connected keyword cluster to
the whole network, ultimately making it a network consisting of article clusters.
This observation further indicates the existence of hierarchical structures in the key-
word networks. Hierarchies in networks mean that, like in the case at hand, some
groups in a network are small and strongly interconnected while other groups are larger
and less interconnected but span across multiple smaller groups (Ravasz & Barabási,
2003). To analyze the hierarchical structure of a network, the correlations between the

TABLE 10.2 Structural network measures

Journal Articles Nodes Edges Avg. Density Avg. Path Clustering

(Keywords) Degree Length Coefficient
CR 592 1449 16,102 22.23 0.015 2.51 0.75
NMS 1235 2953 45,063 30.52 0.010 2.39 0.78
JS 867 2052 25,580 24.93 0.012 2.34 0.79
PC 324 758 7064 18.64 0.025 2.34 0.80
JMP 226 755 5466 14.48 0.019 2.70 0.80
PRR 871 1867 21,628 23.17 0.012 2.33 0.82
MCQ 271 762 5997 15.74 0.021 2.51 0.80
SC 345 896 8165 18.23 0.020 2.40 0.82
Total 4731 6040 116,089 38.44 0.006 2.48 0.76

Note. The measures are based on the unweighted networks and were computed using the NetworkX
package for Python (NetworkX, 2020—v2.5).
 Patterns in communication science 203

degree and the clustering coefficients of the nodes in a network can be investigated
(Choi et al., 2011). If a network is hierarchically structured, nodes with a lower degree
will have a higher clustering coefficient, resulting in a linear asymptotic distribution
connecting the upper end of the clustering scale with the upper end of the degree scale
(Ravasz & Barabási, 2003). In Figure 10.1, the degree-clustering distributions for all

FIGURE 10.1Correlations between clustering coefficient and degree per network.

Note. The x-axes and the y-axes are logarithmically scaled.
204 N. Pfiffner

networks are displayed. All networks follow the expected distribution for hierarchical
networks, which implies that there are some more important keywords that serve as
hub-nodes and connect otherwise distant sub-clusters (Choi et al., 2011). This means
that all networks contain some overarching keywords that play an important role in
uniting the keyword networks and that this is true for the complete aggregated net-
work as well as the journal networks covering separate sub-domains.
To investigate similarities between the networks from a thematic perspective, the
proportion of shared nodes and shared edges, which are reported in Table 10.3 can
be analyzed. Each row reports the overlap regarding the nodes and edges of one
journal with the other journals listed in the columns. The number reported at the
top of each cell is the proportion of shared nodes or edges between journals with
the actual number of overlapping nodes or edges in brackets.
The upper part of Table 10.3 shows significant overlaps in terms of the key-
words in all journals, indicating the existence of common topics among sub-areas
of communication science. Journal networks with many overlaps are potentially
important in serving as a platform to bring different thematical sub-areas together.
As was indicated in Table 10.2, New Media & Society seems particularly relevant, as
all the other journals share a large part of their node-sets with it. This could either
mean that NMS generally covers a large variety of different topics or it could
indicate that topics such as new (digital) media and societal implications are relevant
aspects across all sub-disciplines. Interestingly, although still showing high overlaps
with the other networks, the network for Communication Research (CR), a journal
that has labeled itself as an outlet covering all areas within the field of commu-
nication, is less overlapped by the other networks compared to NMS.
While focusing on the nodes can help to understand the similarities with regards
to individual thematic components, it does not contribute to the understanding of
the composition of these topics. Looking at the shared edges, on the other hand,
can give insights into the distribution of concept relationships among the networks,
and in this regard all networks show a particularly large overlap with the NMS
network, just on a smaller scale. This shows that the sub-domains of communica-
tion science are not distinct from each other; there are common topics shared by
more than one discipline that are then investigated in different constellations
according to sub-field-specific interests.
Furthermore, the overlap measures show some possible hierarchical patterns:
Topics concerning new media technologies and society seem to be present across
all networks, possibly placing these topics at the center of communication science
as a whole. The area of management communication and organizational commu-
nication represented by MCQ, on the other hand, seems to be less important for
other disciplines, while itself incorporates many concepts and structures from other
sub-disciplines. But this should not be overinterpreted because the number of
articles and keywords in the networks is not equally distributed, and the probability
of large networks containing more nodes of other node-sets is most likely higher
than in the case of smaller networks. Still, these findings add to the expectation that
there exists a set of unifying core topics among the journal networks.
TABLE 10.3 Similarities of unweighted journal networks

Node overlap
CR NMS JS PC JMP PRR MCQ SC Avg.
CR 1.0 (1449) 0.551 (798) 0.386 (560) 0.284 (412) 0.312 (452) 0.406 (588) 0.239 (346) 0.276 (400) 0.351 (508)
NMS 0.270 (798) 1.0 (2953) 0.338 (997) 0.166 (489) 0.158 (468) 0.294 (867) 0.148 (438) 0.173 (510) 0.221 (652)
JS 0.273 (560) 0.486 (997) 1.0 (2052) 0.208 (426) 0.160 (329) 0.363 (744) 0.170 (349) 0.205 (420) 0.266 (545)
PC 0.544 (412) 0.645 (489) 0.562 (426) 1.0 (758) 0.323 (245) 0.503 (381) 0.257 (195) 0.334 (253) 0.453 (343)
JMP 0.599 (452) 0.620 (468) 0.436 (329) 0.325 (245) 1.0 (755) 0.433 (327) 0.269 (203) 0.331 (250) 0.430 (324)
PRR 0.315 (588) 0.464 (867) 0.399 (744) 0.204 (381) 0.175 (327) 1.0 (1867) 0.216 (403) 0.238 (444) 0.287 (535)
MCQ 0.454 (346) 0.575 (438) 0.458 (349) 0.256 (195) 0.266 (203) 0.529 (403) 1.0 (762) 0.290 (221) 0.404 (307)
SC 0.446 (400) 0.569 (510) 0.469 (420) 0.282 (253) 0.279 (250) 0.496 (444) 0.247 (221) 1.0 (896) 0.398 (356)
Avg. 0.414 (508) 0.559 (652) 0.435 (545) 0.246 (343) 0.239 (324) 0.432 (535) 0.221 (307) 0.264 (356) —

Edge overlap
CR NMS JS PC JMP PRR MCQ SC Avg.
CR 1.0 (16,102) 0.166 (2670) 0.083 (1330) 0.073 (1182) 0.063 (1017) 0.088 (1409) 0.034 (546) 0.056 (896) 0.080 (1288)
NMS 0.059 (2670) 1.0 (45,063) 0.077 (3453) 0.032 (1453) 0.022 (1010) 0.052 (2331) 0.015 (692) 0.027 (1210) 0.041 (1847)
JS 0.052 (1330) 0.135 (3453) 1.0 (25,580) 0.048 (1217) 0.018 (468) 0.076 (1935) 0.016 (406) 0.038 (981) 0.055 (1406)
PC 0.167 (1182) 0.206 (1453) 0.172 (1217) 1 (7064) 0.049 (343) 0.111 (786) 0.027 (193) 0.068 (481) 0.114 (805)
JMP 0.186 (1017) 0.185 (1010) 0.086 (468) 0.063 (343) 1 (5466) 0.088 (483) 0.029 (156) 0.064 (348) 0.100 (546)
PRR 0.065 (1409) 0.108 (2331) 0.089 (1935) 0.036 (786) 0.022 (483) 1.0 (21628) 0.038 (825) 0.045 (967) 0.058 (1254)
MCQ 0.091 (546) 0.115 (692) 0.068 (406) 0.032 (193) 0.026 (156) 0.138 (825) 1 (5997) 0.048 (289) 0.074 (443)
SC 0.110 (896) 0.148 (1210) 0.120 (981) 0.059 (481) 0.043 (348) 0.118 (967) 0.035 (289) 1 (8165) 0.090 (734)
Avg. 0.104 (1288) 0.152 (1847) 0.099 (1406) 0.049 (805) 0.035 (546) 0.096 (1254) 0.027 (443) 0.049 (734) —

Note. The rows present the fraction of overlap of one journal with the journals mentioned in the columns. For example, the first row of the node overlap shows that 55.1%
Patterns in communication science 205

of the nodes (keywords) of the journal Communication Research (CR) are also part of the node-set of New Media & Society (NMS). The number in brackets represents the
actual number of overlapping nodes or edges. Proportions of 50% overlap or more are marked in bold. The average is computed by summing the row/columns values,
excluding the self-referencing cell, and dividing by 7.
206 N. Pfiffner

The overlap measures in Table 10.3 show the commonalities between pairs of
sub-fields. To further investigate the core commonalities of communication science,
the intersection of the node- and edge-sets of all thematic networks is created. This
results in 77 keywords that are present in all networks, which is less than 1.3% of all
6040 keywords in the complete network. Regarding commonly shared edges, only
27 keyword pairs appear across all the networks. These edges form three isolated
networks, one consisting of 18 nodes and two consisting of 2 nodes each.
Overall, these numbers show that some shared concepts exist that are important
across all analyzed sub-disciplines, although the commonalities might be based on only
a few articles. Furthermore, there are only a few shared edges, which implies that the
common topics are investigated in different constellations across the sub-disciplines.
These observations suggest that the common core is rather weak.

Core concepts in communication science

The structural characteristics presented in the previous section showed that a
common thematical core exists across communication sub-disciplines. Next, a
closer look is taken at the keywords that constitute this core. Figure 10.2 displays
the main edge structure that can be found in all networks and is therefore regarded
as the common core structure of communication science. First, it is clear that three

frame
public
perception

effect identity
communication knowledge

model
online social network

use medium theory

analysis

participation
influence crisis

FIGURE 10.2Common structures across all networks.

Note. Unweighted network structures that appeared in each journal network at least once.
 Patterns in communication science 207

general terms—communication, medium,3 and social—are at the core of the network

of keywords that are shared across all sub-fields. The word communication is still
often used in the macro context as indicated by its links with public, and to some
extent, effect and frame. The word medium, on the other hand, more strongly reflects
the micro context with its links to the words use and influence. Furthermore, the
word medium is also connected to the meso context—its link to crisis either refer-
ences the (economical) media and journalism crisis, or the role of the media in
managing an organizational crisis.
The word social appears in psychological and epistemological contexts (together
with the words perception and knowledge), as well as methodological (together with
the words network and analysis) and political (together with the word participation)
contexts. Finally, four words that connect the three core keywords are public,
identity, online, and theory. These words are particularly important as they reflect the
trending research topics that are at the core of communication studies today as well
as their possible directions to develop: As a result of the growing use of online
media, communication scholars are required to redefine the concepts of public
(macro) and identity (micro), and to reassess their theories accordingly.
While this keyword structure appears in all journals, the graph presented in Figure
10.2 does not tell how prevalent these concepts are in the different networks. It rather
shows that these nodes and edges are common to all networks. To find out more
about the significance of these nodes for the respective networks, looking at the node-
level centrality measures is a valuable approach. The three most basic centrality mea-
sures are degree centrality, betweenness centrality, and closeness centrality. While
degree centrality measures the centrality of a node at the local level, betweenness and
closeness centrality consider the whole network structure. In the context of the study
at hand, these three measures were highly correlated, and the following section will
therefore focus solely on the degree centrality.
Table 10.4 provides an overview of the normalized weighted degree centralities of
the nodes of the core structure in each of the networks. It shows that the importance of
single nodes depends on the thematic sub-field and none of the nodes is equally
important across all areas. General keywords such as medium, social, and communication
show a relatively high importance across all networks, but already here differences can
be observed. While (mass) media is an important topic across most included sub-dis-
ciplines, in management communication (MCQ) the media play a minor role and other
forms of communication seem to be more important. Social aspects seem to be less
important for journalism studies (JS) and science communication (SC) compared to the
other fields. The keyword online has a relatively lower centrality in public relations
(PRR), management communication, and science communication fields, while being
important for new media studies (NMS) and media psychology (JMP). Media psy-
chology also seems to be the most theory-driven field, with the keywords theory and
model being particularly central in its keyword network. These two keywords are also
central in the network for the journal of communication research (CR) where other
psychology-related keywords such as effect and processing have high centralities as well,
reflecting the psychological affinity of this journal. Then there are some nodes that
208 N. Pfiffner

TABLE 10.4 Normalized weighted degrees of core nodes per network

Complete CR NMS JS PC JMP PRR MCQ SC

medium 1.00 0.82 1.00 0.71 1.00 1.00 0.47 0.14 0.50
social 0.79 0.85 0.98 0.18 0.39 0.78 0.51 0.42 0.24
commu- 0.65 1.00 0.35 0.10 0.28 0.49 0.63 0.87 0.95
nication
public 0.53 0.19 0.12 0.23 0.20 0.10 1.00 0.09 0.67
analysis 0.34 0.28 0.22 0.36 0.30 0.13 0.14 0.15 0.35
theory 0.30 0.47 0.18 0.17 0.08 0.51 0.21 0.38 0.17
online 0.29 0.26 0.45 0.16 0.12 0.40 0.05 0.05 0.05
network 0.19 0.17 0.28 0.03 0.16 0.10 0.08 0.13 0.07
crisis 0.15 0.08 0.01 0.06 0.01 0.05 0.34 0.08 0.06
effect 0.15 0.53 0.06 0.04 0.26 0.55 0.03 0.03 0.07
model 0.15 0.32 0.08 0.06 0.12 0.19 0.09 0.11 0.12
identity 0.13 0.18 0.12 0.07 0.05 0.27 0.04 0.31 0.04
frame 0.13 0.12 0.03 0.11 0.19 0.18 0.08 0.04 0.24
use 0.12 0.14 0.16 0.04 0.06 0.32 0.02 0.05 0.04
participa- 0.11 0.14 0.16 0.03 0.12 0.06 0.02 0.04 0.05
tion
perception 0.08 0.10 0.02 0.04 0.03 0.17 0.05 0.02 0.26
knowledge 0.08 0.13 0.05 0.04 0.12 0.07 0.01 0.06 0.16
influence 0.06 0.18 0.02 0.03 0.06 0.12 0.02 0.02 0.05

Note. Degrees were weighted by dividing node degree by the maximum degree in the network.

dominate one field but are less important for all the others. For example, crisis has a very
central position in the public relations network, while its centrality scores are quite low
in all other networks.
Lastly, these observations can further be compared with the list of the most
important keywords in the networks. Table 10.5 lists the 15 most important key-
words for each network according to their weighted degree centrality. This over-
view shows that although in all networks some of the core structure’s keywords
appear among the most central ones (highlighted in bold) they represent rather
general concepts. Furthermore, it becomes obvious that each sub-field is char-
acterized by its unique set of important concepts. In summary, it can therefore be
concluded that, based on this analysis, there is a common core structure that
appears in all sub-disciplines that consists of general concepts and that each sub-
discipline can be further characterized based on its unique set of central keywords.

Discussion
The objective of this study was to investigate what the structural commonalities
among communication science sub-domains are (RQ1), and to identify the most
TABLE 10.5 Top keywords with the highest weighted degree centrality per network

Complete CR NMS JS PC JMP PRR MCQ SC

Medium 0.43 commu- 0.39 medium 0.51 journalism 0.58 medium 0.54 medium 0.30 public 0.64 organiza- 0.46 science 0.49
nication tional
Social 0.37 Social 0.34 social 0.49 news 0.48 political 0.50 social 0.24 relation 0.59 commu- 0.41 commu- 0.49
nication nication
commu- 0.33 Medium 0.33 digital 0.35 medium 0.44 news 0.33 effect 0.19 commu- 0.41 theory 0.22 public 0.34
nication nication
Public 0.28 Theory 0.24 online 0.30 analysis 0.25 social 0.24 theory 0.18 social 0.34 organization 0.22 medium 0.29
News 0.23 Effect 0.24 internet 0.29 newspaper 0.21 commu- 0.23 video 0.17 medium 0.32 social 0.21 climate 0.25
nication
Journalism 0.23 News 0.21 commu- 0.25 public 0.19 campaign 0.22 game 0.17 crisis 0.25 manage- 0.19 change 0.24
nication ment
Theory 0.21 Model 0.18 technol- 0.21 journalistic 0.16 analysis 0.22 commu- 0.17 theory 0.19 identity 0.17 analysis 0.21
ogy nication
Analysis 0.21 Information 0.17 network 0.20 social 0.16 election 0.20 online 0.15 management 0.19 work 0.15 social 0.19
Relation 0.20 Political 0.15 analysis 0.17 political 0.16 effect 0.18 emotion 0.14 corporate 0.18 employee 0.13 engagement 0.19
Digital 0.20 Analysis 0.15 mobile 0.16 content 0.16 frame 0.16 news 0.13 relationship 0.16 discourse 0.13 risk 0.18
Online 0.20 Message 0.14 theory 0.15 theory 0.15 politic 0.16 enjoyment 0.12 strategy 0.14 leadership 0.13 frame 0.16
Political 0.17 Online 0.13 political 0.15 press 0.15 information 0.16 use 0.11 analysis 0.14 relationship 0.11 perception 0.16
Internet 0.16 computer- 0.13 game 0.15 discourse 0.14 public 0.16 identity 0.10 pr 0.14 medium 0.09 environ- 0.15
mediate mental
Network 0.14 Attitude 0.12 culture 0.15 online 0.13 experiment 0.15 entertainment 0.10 responsibility 0.12 analysis 0.09 research 0.14
Informa- 0.14 Exposure 0.12 commu- 0.14 ethic 0.13 opinion 0.14 motivation 0.08 image 0.12 rhetoric 0.09 knowledge 0.14
tion nity
Model 0.14 Processing 0.12 facebook 0.14 journalist 0.12 content 0.13 model 0.08 model 0.11 technology 0.09 theory 0.13
Patterns in communication science 209
210 N. Pfiffner

important topics across all sub-domains (RQ2) based on a semantic network ana-
lysis of keywords in research articles published between 2008 and 2019 in eight
communication journals, each devoted to a specific sub-domain of the field.
Communication science is often characterized as a fragmented research field
(Waisbord, 2019) that lacks a common identity and is split into sub-domains that show
little interconnections among and interest in each other (Craig, 1999; Donsbach,
2006). Empirical studies investigating the structural commonalities in communication
science are rare and the focus often lies on differences rather than commonalities
among disciplines. In the study at hand, the aim was therefore to detect commonalities
among the sub-disciplines of communication science. It was found that among the
journals included, the overlap in terms of keyword edges (pairs) was relatively low,
ranging from 1.5% to about 16%. This reflects the fragmentation within the discipline.
New Media & Society, which is the leading journal of communication in terms of its
impact factor, was identified as the journal that shares the most nodes and edges with
the other journals, followed by Communication Research. Both journals cover a rela-
tively wide thematic spectrum. This finding is in line with the observations by Barnett
et al. (2011), who found that general communication journals unite the field by
having ties to many other (sub-)disciplines. Similarly, Günther and Domahidi (2017)
also found that general communication journals cover a wide variety of topics.
With regards to the common core of communication science (RQ1), the cur-
rent study found that 77 keywords of 6040 keywords in the complete network are
shared among all journals. These keywords display the still important division
between macro- and micro-level studies of communication, as well as the growing
focus on the online media and its implications on changes in the public sphere as
well as in the formation of individual identity. The centrality of the word theory in
this context is particularly interesting as it seems that communication scholars are
still in search of relevant theories to explain those transformations.
These common structures then served as the basis to investigate the most important
topics across all sub-domains (RQ2). Three larger thematical sub-clusters that appear in
all journal networks were identified: Online communication, psychological aspects (in parti-
cular media influence and media effects), and social relationships and interpersonal processes.
Other studies have also recognized the importance of online communication as an
important driver of communication science that has, on the one hand, become a central
part of the research field (Günther & Domahidi, 2017; Montero-Díaz et al., 2018;
Tomasello et al., 2010), and on the other hand has brought together the formerly often
distinct fields of mass and interpersonal communication (O’Sullivan & Carr, 2018).
Relating to this, the shared interest in social relationships and interpersonal processes
might be linked to the common interest in online communication, since these two
clusters are directly connected in the core structure. Furthermore, the interest in the
psychological aspects of communication and the influence of psychology as a discipline
has long been acknowledged (Barnett et al., 2011; Park & Leydesdorff, 2009; So, 1988).
Next to these three larger thematic clusters, framing theory as well as the keyword triad
of social–network–analysis appeared across all studied networks. Framing theory has
been identified before as the most prominent theoretical approach in communication
 Patterns in communication science 211

science (Chung et al., 2013; Khan et al., 2016), and the prominence of social network
analysis is most likely linked to the growing interest in online communication and in
new methods to understand it. Overall, the thematic commonalities identified here are
therefore in line with previous findings and hold no surprises.
Based on these findings we can conclude that there exists a common core of com-
munication science, but at the same time it must be noted that the identified core might
be rather volatile as each node and edge only had to appear at least once in all networks
to be included. Furthermore, the common keywords with the highest centrality among
all sub-disciplines such as communication, social, or medium are rather general and do not
convey much meaning by themselves. Consequently, the core or identity of commu-
nication science must be characterized as vague and leaves room for much interpreta-
tion. At the same time, a conclusive interpretation of the presented findings in this
regard is difficult, because what qualifies as a strong or weak common keyword or
knowledge structure in the case of a scientific domain remains an open question. To be
able to interpret the results more accurately, one strategy would therefore be to integrate
non-communication journals into future analyses to explicitly assess if the links and
commonalities between communication sub-disciplines are unique.
Lastly, some limitations of this study must be acknowledged. First, the selection of
journals representing the different sub-disciplines was limited by the availability of
article keywords and therefore some of the larger communication journals were not
included in the analysis. Although the focus of this study was not to identify the gen-
eral structure of communication science but rather uncover the commonalities across
sub-disciplines, it would have still been interesting to incorporate other major com-
munication journals in the analysis. Furthermore, journals focusing explicitly on other
sub-disciplines such as interpersonal communication would have been interesting to
include but have been left out due to missing article keywords. Future studies could
therefore focus on identifying the commonalities across communication science sub-
disciplines using other publication features such as the title, or automatically extracted
topics based on the abstracts or the full texts to be more flexible in the sample com-
position and avoid relying on the availability of keywords.

Tips and lessons from this chapter

1. Looking at the structural parameters of the network can already provide

valuable insights. It is especially helpful to compare different networks before
conducting a deeper thematical analysis. In this chapter, we looked more
closely at the network density and the (average) clustering coefficient. Network
density is a common measurement, where high density indicates that the
concepts included in the semantic network are more strongly connected to
each other, making the network more homogenous compared to a network
with lower density. The average clustering coefficient describes the tendency of
nodes in the network to cluster together. When this coefficient is high, the
network consists of many tightly connected clusters, meaning that many
keywords form relatively independent sub-groups.
212 N. Pfiffner

2. In the study at hand, the compared networks did not show large variations
with regards to these measures, which indicates that they are structurally
similar. This does not come as a surprise, as the networks were derived from
the same institutionalized settings (journals) that all essentially follow the
same logic. Still, with regards to the density we could already see that the
journal Political Communication had a narrower thematic focus than New
Media & Society, as the latter had a significantly lower density. This observa-
tion was later supported by the examination of overlapping edges among
the journals reported in Table 10.3.
3. If the journal networks would have shown significantly different structural
parameters, it would have been appropriate to take a step back and
examine these differences more closely before moving to the interpreta-
tion of the concepts underlying the network nodes.
4. It is often useful to compare different structural network measures. For
example, in this chapter we looked at the distribution of node degrees,
conditional on their clustering coefficient. This comparison helped to
uncover the hierarchical structures of the investigated networks. A variety
of measurements limits the biases of misinterpretation and opens new
possibilities for additional insights into the network structure.
5. It is important to consider the nature of the words analyzed in the context of a
semantic network analysis. Keywords chosen by domain experts convey a
different meaning than those extracted through automatic content analysis.
Their link to each other is not based on their co-occurrence in the same sen-
tence, but rather their association with the article itself and the extent to
which these keywords appeal to the academic community. With the choice of
keywords, authors of academic papers attempt, among other things, to pro-
mote their articles and make them accessible. In a way, keywords function as
hashtags, tags to images, or other folksonomies in social media. The inter-
pretation of the networks should therefore take this context into account.
6. When assessing the similarities or differences of networks, it is often hard to
interpret the findings on their own. Therefore, it is helpful to look for similar
studies and use their results as a benchmark to compare your own results to
and get some indications about the magnitude of your findings. Of course,
sometimes it is hard to find related studies that are suitable for comparison,
but even distantly related findings can serve as an interpretation aid.

Notes
1 This list included: “”, “a”, “all”, “an”, “and”, “as”, “at”, “for”, “in”, “of”, “on”, “or”,
“than”, “that”, “the”, “then”, “this”, and “to”.
2 For the lemmatization, the lemmatizer implemented in the spaCy package (spaCy 2020—
v2.3.4) for Python was used.
3 Note that the node “medium” also stands for “media” as during the lemmatization pro-
cess “media” was converted to “medium”.
 Patterns in communication science 213

References
ALAIC. (2020). Grupos de Temáticos (GT). Asociación de Investigatores de la Comuni-
caión. https://www.alaic.org/site/grupos-de-trabalho/.
Assefa, S. G., & Rorissa, A. (2013). A bibliometric mapping of the structure of STEM
education using co-word analysis. Journal of the American Society for Information Science and
Technology, 64(12), 2513–2536. doi:10.1002/asi.22917.
Barnett, G. A., & Danowski, J. A. (1992). The Structure of Communication. Human Com-
munication Research, 19(2), 264–285. doi:10.1111/j.1468-2958.1992.tb00302.x.
Barnett, G. A., Huh, C., Kim, Y., & Park, H. W. (2011). Citations among communication
journals and other disciplines: a network analysis. Scientometrics, 88(2), 449–469.
doi:10.1007/s11192-011-0381-2.
Berger, C. R., Roloff, M. E., & Roskos-Ewoldsen, D. R. (2010). What is communication
science? In C. R. Berger, M. E. Roloff, & D. R. Roskos-Ewoldsen (Eds.), The Handbook
of Communication Science (pp. 3–20). New York: SAGE Publications. doi:10.4135/
9781412982818.n1.
Borah, P. (2017). Emerging communication technology research: Theoretical and metho-
dological variables in the last 16 years and future directions. New Media & Society, 19(4),
616–636. doi:10.1177/1461444815621512.
Boyack, K. W., & Klavans, R. (2010). Co-citation analysis, bibliographic coupling, and
direct citation: Which citation approach represents the research front most accurately?
Journal of the American Society for Information Science and Technology, 61(12), 2389–2404. http
s://doi.org/10.1002/asi.21419.
Bryant, J., & Pribanic-Smith, E. (2010). A historical overview of research in communication
science. In C. R. Berger, M. E. Roloff, & D. R. Roskos-Ewoldsen (Eds.), The Handbook of
Communication Science (pp. 21–36). New York: SAGE Publications. doi:10.4135/
9781412982818.n2.
Choi, J., Yi, S., & Lee, K. C. (2011). Analysis of keyword networks in MIS research and
implications for predicting knowledge evolution. Information & Management, 48(8), 371–381.
doi:10.1016/j.im.2011.09.004.
Chung, C. J., Barnett, G. A., Kim, K., & Lackaff, D. (2013). An analysis on communication
theory and discipline. Scientometrics, 95(3), 985–1002. doi:10.1007/s11192-012-0869-4.
Chung, C. J., Lee, S., Barnett, G. A., & Kim, J. H. (2009). A comparative network analysis
of the Korean Society of Journalism and Communication Studies (KSJCS) and the
International Communication Association (ICA) in the era of hybridization. Asian Journal
of Communication, 19(2), 170–191. doi:10.1080/01292980902827003.
Craig, R. T. (1993). Why Are There So Many Communication Theories? Journal of Com-
munication, 43(3), 26–33. doi:10.1111/j.1460-2466.1993.tb01273.x.
Craig, R. T. (1999). Communication theory as a field. Communication Theory, 9(2), 119–161.
doi:10.1111/j.1468-2885.1999.tb00355.x.
Crossref. (2020). Crossref REST API [Computer software]. Crossref. https://github.com/
CrossRef/rest-api-doc.
Delia, J. G. (1989). Communication research: A history. In C. R. Berger & S. H. Chaffee
(Eds.), Handbook of communication science (pp. 20–98). New York: SAGE Publications.
Ding, Y., Chowdhury, G. G., & Foo, S. (2001). Bibliometric cartography of information
retrieval research by using co-word analysis. Information Processing & Management, 37(6),
817–842. doi:10.1016/s0306-4573(00)00051–0.
Donsbach, W. (2006). The Identity of Communication Research. Journal of Communication,
56(3), 437–448. doi:10.1111/j.1460-2466.2006.00294.x.
214 N. Pfiffner

ECREA. (2020). Sections. European Communication Research and Education Association.

https://ecrea.eu/Sections.
Evans, J. A., & Foster, J. G. (2011). Metaknowledge. Science, 331(6018), 721–725.
doi:10.1126/science.1201765.
Günther, E., & Domahidi, E. (2017). What Communication Scholars Write About: An
Analysis of 80 Years of Research in High-Impact Journals. International Journal of Commu-
nication, 11, 3051–3071.
Gupta, B. M., & Bhattacharya, S. (2004). Bibliometric approach towards mapping the
dynamics of science and technology. DESIDOC Bulletin of Information Technology, 24(1),
3–8. doi:10.14429/dbit.24.1.3616.
Hanneman, R. A., & Riddle, M. (2011). Concepts and Measures for Basic Network Ana-
lysis. In P. J. Carrington & J. Scott (Eds.), The SAGE Handbook of Social Network Analysis
(pp. 340–369). New York: SAGE Publications.
ICA. (2020). Homepage. International Communication Association. https://www.icahdq.org/.
Ignatow, G., & Mihalcea, R. (2017). Text Mining: A Guidebook for the Social Sciences. New
York: SAGE Publications. doi:10.4135/9781483399782.
Jiménez, L. G., & Guillem, S. M. (2012). Does communication studies have an identity?
Setting the bases for contemporary research. Catalan Journal of Communication & Cultural
Studies, 1(1), 15–27. doi:10.1386/cjcs.1.1.15/1.
Kamhawi, R., & Weaver, D. (2003). Mass Communication Research Trends from 1980 to
1999. Journalism & Mass Communication Quarterly, 80(1), 7–27. doi:10.1177/
107769900308000102.
Khan, G. F., Lee, S., Park, J. Y., & Park, H. W. (2016). Theories in communication science:
a structural analysis using webometrics and social network approach. Scientometrics, 108(2),
531–557. doi:10.1007/s11192-015-1822-0.
Kim, M. C., & Zhu, Y. (2018). Scientometrics of scientometrics: Mapping historical foot-
print and emerging technologies in scientometrics. In M. Jibu & Y. Osabe (Eds.), Scien-
tometrics. InTech. doi:10.5772/intechopen.77951.
Koivisto, J., Thomas, P. D., & Aslama, M. (2010). Mapping communication and media research:
Conjunctures, institutions, challenges. Tampere: Tampere University Press.
Lee, K., Jung, H., & Song, M. (2016). Subject–method topic network analysis in commu-
nication studies. Scientometrics, 109(3), 1761–1787. doi:10.1007/s11192-016-2135-7.
Leydesdorff, L., & Milojević, S. (2015). Scientometrics. In J. D. Wright (Ed.), International
Encyclopedia of the Social & Behavioral Sciences (pp. 322–327). Leiden: Elsevier. doi:10.1016/
B978-0-08-097086-8.85030-8.
Littlejohn, S., & Foss, K. (2009). Encyclopedia of Communication Theory. SAGE Publications,
Inc. doi: 10.4135/9781412959384.
Luke, D. A. (2015). A user’s guide to network analysis in R. Use R!New York: Springer.
McCroskey, J. C., & Richmond, V. P. (1996). Human Communication Theory and
Research: Traditions and Models. In M. B. Salwen & D. W. Stacks (Eds.), LEA’s com-
munication series. An integrated approach to communication theory and research (pp. 233–242).
Boca Raton: Erlbaum.
McMahan, D. T. (2004). What we have here is a failure to communicate: linking interpersonal
communication and mass communication. Review of Communication, 4(1–2),33–56.
doi:10.1080/1535859042000250290.
Montero-Díaz, J., Cobo, M.‑J., Gutiérrez-Salcedo, M., Segado-Boj, F., & Herrera-
Viedma, E. (2018). A science mapping analysis of “Communication” WoS subject
category (1980–2013). Comunicar, 26(55), 81–91. doi:10.3916/C55-2018-08.
NetworkX. (2020). NetworkX (Version 2.5) [Computer software]. NetworkX. https://net
workx.org/documentation/stable/index.html.
 Patterns in communication science 215

O’Sullivan, P. B., & Carr, C. T. (2018). Mass personal communication: A model bridging
the mass-interpersonal divide. New Media & Society, 20(3), 1161–1180. doi:10.1177/
1461444816686104.
Park, H. W., & Leydesdorff, L. (2009). Knowledge linkage structures in communication
studies using citation analysis among communication journals. Scientometrics, 81(1), 157–
175. doi:10.1007/s11192-009-2119-y.
Potter, W. J., & Riddle, K. (2007). A Content Analysis of the Media Effects Literature. Journal-
ism & Mass Communication Quarterly, 84(1), 90–104. doi:10.1177/107769900708400107.
Ravasz, E., & Barabási, A.‑L. (2003). Hierarchical organization in complex networks. Phy-
sical Review. E, Statistical, Nonlinear, and Soft Matter Physics, 67(2), 26112. doi:10.1103/
PhysRevE.67.026112.
Reardon, K. K., & Rogers, E. M. (1988). Interpersonal Versus Mass Media Communication
A False Dichotomy. Human Communication Research, 15(2), 284–303. doi:10.1111/j.1468-
2958.1988.tb00185.x.
Riffe, D., & Freitag, A. (1997). A Content Analysis of Content Analyses: Twenty-Five Years
of Journalism Quarterly. Journalism & Mass Communication Quarterly, 74(4), 873–882.
doi:10.1177/107769909707400414.
Rogers, E. M. (1999). Anatomy of the Two Sub-disciplines of Communication Study. Human
Communication Research, 25(4), 618–631. doi:10.1111/j.1468-2958.1999.tb00465.x.
Shaw, D. L., Hamm, B. J., & Knott, D. L. (2000). Technological Change, Agenda Chal-
lenge and Social Melding: mass media studies and the four ages of place, class, mass and
space. Journalism Studies, 1(1), 57–79. doi:10.1080/146167000361177.
Simonson, P. (2015). Communication and Media Studies, History since 1968. In W.
Donsbach (Ed.), The International Encyclopedia of Communication (pp. 1–8). London: John
Wiley & Sons, Ltd. doi:10.1002/9781405186407.wbiecc086.pub3.
So, C. Y. K. (1988). Citation Patterns of Core Communication Journals: An Assessment of
the Developmental Status of Communication. Human Communication Research, 15(2),
236–255. doi:10.1111/j.1468-2958.1988.tb00183.x.
spaCy. (2020). spaCy (Version 2.3.4) [Computer software]. https://spacy.io/.
Tomasello, T. K., Lee, Y., & Baer, A. P. (2010). “New media” research publication trends
and outlets in communication, 1990–2006. New Media & Society, 12(4), 531–548.
doi:10.1177/1461444809342762.
Trumbo, C. W. (2004). Research Methods in Mass Communication Research: A Census of
Eight Journals 1990–2000. Journalism & Mass Communication Quarterly, 81(2), 417–436.
doi:10.1177/107769900408100212.
Waisbord, S. R. (2019). Communication: A Post-Discipline. London: Polity.
Waltman, L., van Eck, N. J., & Noyons, E. C. (2010). A unified approach to mapping and
clustering of bibliometric networks. Journal of Informetrics, 4(4), 629–635. doi:10.1016/j.
joi.2010.07.002.
Wartella, E. (1996). The history reconsidered. In E. E. Dennis & E. Wartella (Eds.), LEA’s
Communication Series. American Communication Research: The Remembered History (pp. 169–
180). Boca Raton: Erlbaum.
Yang, H., & Jung, W.‑S. (2016). Structural dynamics of keyword networks: Liquid crystal
display and plasma display panel cases. Journal of Engineering and Technology Management, 40,
64–75. doi:10.1016/j.jengtecman.2016.04.002.
11
SUMMARY AND CONCLUSION
Elad Segev

Words are the building blocks of communication, and communication is the basis
of our society. Throughout history we have put together words to communicate
our needs, coordinate our actions, and craft our stories, in order to organize
increasingly larger groups, support each other, compete for resources, and more
generally survive and leave our mark on humanity. Thus, words entail a lot of
power. Yet words are also a fantastic resource for scholars, historians, sociologists,
political, and communication scientists, who attempt to trace back the nature of
people, their cultures, beliefs, histories, and power structures.
The tremendous value of text was acknowledged by researchers over the years
who proposed different methods and approaches to analyze it, from the studies on
sacred texts and the development of logic and rhetoric in ancient times to the more
recent methods of discourse analysis, framing, structuralism, and post-structuralism.
What is common to all these methods is the need to break the text into smaller
analytical units and then decompose it again to extract its meanings and purposes.
Semantic network analysis is a relatively new approach in the field. It uses
computational and visualization methods and enables big data analyses that were
not possible before. The growing amount of information available today makes this
method even more appealing and crucial for researchers.
In this book we first introduced the idea behind semantic network analysis, its
premises, historical development, and current place, among other computational
research methods. We then explained in detail the stages and tools used to conduct
semantic network analysis. Finally, much of the book focused on the applications
of semantic network analysis in different types of texts, research settings, and dis-
ciplines. This was divided into three parts: semantic network analysis in a top-
down text (such as news articles or political speeches), in bottom-up texts (such as
social media messages), and in research data (such as academic papers and interview
transcripts).
DOI: 10.4324/9781003120100-12
 Summary and conclusion 217

The book first presented the applications of semantic network analysis from top-
down texts. Noa Hatzir, Kohei Watanabe, Atsushi Tago, and I (Elad Segev) in
Chapter 2 analyzed Israeli news related to nuclear threats. The method in this case
helped us to identify the main actors in the story and the common topics and
issues. We could even draw on the meta-narrative presented in the Israeli media—
the fear from war, which appeared at the center of all network maps. At the same
time, semantic network analysis also revealed the differences between news outlets
with different political inclinations. The news was also the primary source for the
study in Chapter 3, in which Yuan Zhou and Sandrine Boudana and I focused on
the mentions of the word provocation to explore provocation narratives and their
political biases. In this research, semantic networks allowed us, in line with the
post-structuralist approach, to show the hidden power structures in international
news. As we compared American and Chinese news, we identified major differ-
ences between the two. To our great surprise English and Chinese news did not
contradict one another. It appeared that Chinese news adopted the American nar-
rative about the problematic role of North Korea, but sophisticatedly managed to
shift the responsibility toward the US.
A different kind of news was utilized in the study of James A. Danowski, Bei
Yang, and Ken Riopelle in Chapter 4. Instead of newspapers, their study utilized
television transcripts. They compared how three television channels: CNN, Fox
News, and MSNBC covered the topic of “social distancing” during the COVID-19
pandemic. Semantic network analysis enabled them to examine more closely and
accurately the sentiment related to social distancing and its relation to the market
share growth and the political ideology of each channel. They found that more than
ideology, the market share growth of the channels was a better predictor of the
extent of positive and negative coverage around social distancing.
Chapter 5 demonstrated the use of semantic network analysis in political spee-
ches. Maya Hadar, Regula Miesch and I compared political speeches in Israel and
Germany that mentioned the Holocaust over the last decade. The clusters of words
in each network clearly represented the national narratives and the important
identities of Israel as a country that needs to protect itself from the surrounding
enemies, and of Germany as a country that promotes freedom, tolerance, and
democracy around the world. In this study, semantic network analysis helped to
unveil the political rhetoric and the main narratives of prominent leaders in Israel
and Germany, and to show how the Holocaust is used as an instrument to promote
political goals. This chapter neatly corresponded with the second chapter on the
Israeli coverage of nuclear threats. Holocaust was a prominent word in the Iranian
nuclear map based on news, while the Iranian nuclear plan was one of the main
clusters and themes of the Holocaust map based on political speeches.
The focus on bottom-up texts such a social media content was demonstrated
first in Chapter 6. Jérôme Chariatte and Diana Ingenhoff applied semantic network
analysis to investigate the discourse around Brexit on Twitter and compared it with
the same discourse in international news. They stressed the importance of sub-state
actors such as cities (e.g., London) or countries (e.g., Scotland) in producing the
218 E. Segev

social media discourse, which is known as “paradiplomacy”. They found that this
discourse was much more emotional and polarized for and against Brexit. International
news, on the other hand, was mainly critical\more informative, and focused on the
national goals of the reporting countries. Semantic network analysis in this unique
comparison enabled the mapping of the discourse, highlighting the function and
characteristic of each medium (social media vs. traditional news), and displaying the
rising influence of paradiplomacy and sub-state actors. Their findings have various
applications for politicians and public diplomacy practitioners who wish to commu-
nicate their messages effectively and engage with an increasingly diverse audience.
Another study that used tweets as the source of text was presented in Chapter 7
by Saki Mizoroki and Bumsoo Kim. They utilized semantic networks to compare
the discourse on two widely publicized rape cases of women in Japan and South
Korea. They found that the discourse on Twitter hardly focused on gender-based
violence, but rather quickly diverted into a political debate on national scandals.
While South Korean users demonstrated some female solidarity, it was completely
missing in Japan. Hence, semantic network analysis was found to contribute also to
the literature on feminist activism and textual analysis of tweets.
The last part of the book focused on research data. Noa Hatzir showed in Chapter
8 how to use semantic network analysis to structure and extract reoccurring themes
in interviews transcripts. She studied the perceived and actual role of WhatsApp as a
tool among women in the family. She also compared these perceptions with the
actual content that users post in WhatsApp groups. Her findings highlighted the
advantages of WhatsApp as an asynchronous channel, which helps people to control
their time and facilitate meetings or phone conversations. The actual content of
WhatsApp conversations revealed the importance of phatic communication with its
positivity bias. Hence, semantic network analysis was useful also in comparing dif-
ferent types of texts to get a more complete picture of the phenomenon.
Another application of semantic network analysis in research data was shown in
Chapter 9. Audrey Addi-Raccah, Tali Shahrabani, and I reviewed and mapped the
recent literature on school improvement. We found that the studies of success in
education are influenced by management scholarship. Most studies emphasize the
responsibility of the teachers and school principals, while the students, parents, and
other external stakeholders are viewed as more passive in the process. School
improvement was primarily defined and assessed based on students’ academic abilities
and their outcomes or scores in (particularly math) tests. Here semantic network
analysis helped us to map a sub-field of research and show its lacuna.
Finally, a similar but more general attempt to map the field was done in Chapter
10. Instead of the entire text of the papers, Nico A. Pfiffner focused on author
assigned keywords in the broad field of communication science. Semantic net-
works in this case helped him to explore what topics are studied today in the field
of communication, what the common issues are among all topics, and to what
extent each topic is coherent and unique. Here too, this approach was found to be
powerful in mapping an entire field of study, and in showing its core agendas and
the most important sub-fields.
 Summary and conclusion 219

The network figure on the book cover represents the semantic network of this
book, including all chapters, even the very last one that you read now.1 The colors
of words are obtained based on a cluster analysis (Louvain modularity), indicating
that most of the chapters have a unique focus, but some share a similar scope.
Obviously, at the absolute center of the network are the words “semantic network
analysis”, which correspond to the book title and are common to all studies
included in the book. Connected to the word “analysis” are several words related
to methods such as “qualitative”, “data”, “text”, and “articles”, which are discussed
in all chapters, but particularly in the introduction and conclusion.
Some general disciplinary words, such as “news”, “social media”, “communica-
tion”, and “political”, are still very central in the network, being common to several
studies. News is the basic corpus for the studies in the first part of the book, such as the
study on provocation narratives in US and Chinese news (Chapter 3), or the social
distancing discourse on CNN and Fox news (Chapter 4). Social media is the common
ground for the Twitter studies on Brexit (Chapter 6) and sexual abuse discourse
(Chapter 7). Communication is the central concept in the study of WhatsApp use in
the family (Chapter 8) and, of course, the attempt to map the field of communication
(Chapter 10). Finally, the word “political” is the common ground for the studies on
the coverage of the Iranian nuclear program (Chapter 2) and the Holocaust speeches
of political leaders (Chapter 5).
Hence, without actually “knowing” this book’s structure, the semantic network of
words appearing together in the same sentence allows us to get an overview, a
descriptive map of the content, and the thematic relationships between the chapters.

The key features of semantic network analysis

One of the main advantages of semantic network analysis is that it is simple and
straightforward. It is available to researchers and can be done with relatively little
training. Many of the studies presented in this book came out as part of workshops
conducted on semantic network analysis in Israel, Switzerland, and Japan. Another
advantage as was mentioned in the introduction is that semantic networks reflect
the actual relationships between words as they appear in the text. If done properly,
researchers do not have to impose their own assumptions and prepositions (com-
pared to codebooks, questionnaires, or even qualitative content analysis). Indeed,
there are critical stages in the process, such as the selection of words to include in
the networks or the interpretations of the networks, which require more of the
researcher’s intervention. Still, the beauty of this method is that the findings reflect
closely the text itself, allowing researchers to learn new things about the text
through identifying central words, prominent connections between words, and
clusters of words that were not observed previously.
At the end of each study there is a common section on tips and lessons for the
use of semantic network analysis. Many of the studies in this section highlighted
the advantages of semantic network analysis. Common suggestions that appeared in
most of the studies were related to the power of this method when analyzing large
220 E. Segev

texts. In addition, texts in any given language can be analyzed. At the same time,
most of the studies mentioned the importance of carefully considering the selection
of words to include, the need to sparsify the network to make it readable and
interpretable, and the significance of contextualizing the words.
The power to analyze large texts. In most of the studies presented in the book the
amount of text used for the analysis was relatively large: Tens of hours of interview
transcripts, a few hundreds or thousands of academic papers, tens of thousands of
news items, and hundreds of thousands of tweets. In many cases, rather than a
sample, these data represented the entire text in question (e.g., all the news items
that mentioned a specific country or concept, all the tweets that mentioned a
specific person, all the academic papers in a specific journal). They might not
strictly adhere to the definition of Big Data studies (Kitchin & McArdle, 2016);
nonetheless, these data cannot be processed and analyzed entirely using traditional
manual methods such as codebooks or qualitative methods alone. Semantic network
analysis works well with larger datasets and Big Data studies. In a world of informa-
tion overload, the biggest advantage of semantic networks is that they offer a
descriptive map: Where to start the analysis and what issues to focus on. Still, when it
comes to understanding the main themes in the text, Big Data studies are not always
necessary, and often rather consume unnecessary computational resources and
advanced computation skills. A large text and even a random sample of the text that
contains a few thousand units would often be sufficient (Segev, 2020). For smaller
texts, that include less than a few thousand words, sentiment network analysis might
be less useful, as there are not enough co-occurrences of word pairs to generate
meaningful networks. In those cases, the more traditional content analysis methods
would be more appropriate.
Appropriate for all languages. Semantic network analysis is a suitable approach for any
language. Still, some languages may require more preparation than others. The need
to tokenize the text into words is not always easy. Some languages such as Chinese,
Japanese, or Korean have no spaces between words and therefore require special text
segmentation software to properly identify single words. The studies presented in the
book used the jiebaR algorithm (Qin & Wu, 2019) for Chinese, and KH Coder
(Higuchi, 2016; Higuchi, 2017) for Japanese and Korean. Further, when looking at
the list of the most frequent words, some words might need to be merged as they
convey a very similar or even identical meaning. Thus, in some languages, such as
Hebrew or Arabic, the forms of the words may slightly change depending on the
pronouns and prepositions. The processes of merging similar words into one entity
are called stemming and lemmatization, and they make the network analysis and
interpretation much easier if done at the early stages of text processing.
Careful selection of words. Although semantic network analysis is relatively
straightforward, there are a few stages where researchers’ involvement is critical.
This is particularly important when choosing what text to analyze and what words
to look at when constructing the networks. Many of the studies in the book
pointed at the importance of choosing the search term to use when retrieving the
text and focusing on specific sources. Some even recommended looking only at
 Summary and conclusion 221

sentences that mentioned the desired search term. In some cases, this can be rela-
tively simple, such as when looking at the way news reports on a certain country
(Chapter 2) or the way Twitter users write about a certain person (Chapter 7). In
other cases, for example, when looking at the issue of success in school (Chapter
9), researchers are required to think carefully about the possible search terms that
can best reflect the topic in question. Once the text is defined, another critical
stage is to come up with a meaningful list of words to include in the network.
Here many studies chose to look at frequent words, but some further focused on
names of specific actors, country names, or even keywords of academic papers.
When choosing appropriate words among a list of frequent words, it is often useful
to have a discussion among researchers and look at the context of the words in the
text itself, in order to make sure that they are distinctive and relevant for the study.
Finally, in the networks itself, some studies found it useful to visually separate
between different types of words, such as the actors and their actions in the study
on the Iranian nuclear threat (Chapter 2), or the involving users and the actual
words that they post in WhatsApp groups (Chapter 8).
Sparsifying the network. The first semantic networks produced based on the list of
chosen words are almost always very dense. Another challenge that all studies had
to deal with was the need to sparsify the networks to see their structure and iden-
tify their clusters. In this book we presented common techniques to sparsify the
network, most importantly to remove links that are less frequent as well as to
delete the most central words that were used as search terms to obtain the data. For
example, when looking at news that mentioned the word “provocation” (Chapter
3) the word “provocation” appeared at the absolute center, making it difficult to
identify clusters in the network. Since the word “provocation” was linked to most
other words and the network itself tells the story of provocation narratives in the
news, it is safe to remove this word. The process of sparsification requires keeping a
good balance. When removing too little the network can still be very messy and
the interpretation would be challenging. When removing too much, on the other
hand, the network becomes very dull and hardly offers meaningful information.
Contextualizing the words. The art of network interpretation is the most important
and innovative stage. All studies indicated that it is crucial for the researchers to
have a good knowledge of the text so that the central words and clusters in the
network could be properly contextualized. For this purpose, researchers need to go
back to the text and see the context in which these words and clusters appear, offer
quotes of relevant sentences or posts, and even complement the semantic networks
with qualitative analysis to offer a richer and deeper layer of meaning.

How to interpret the networks?

Although the book does not offer a systematic procedure for the interpretation of
the networks, there are some basic cues to consider. The network serves as a map
that can help researchers to navigate within the text itself. It offers several cues for
how to navigate back and forth from the network to the text.
222 E. Segev

Inspect the most central words in the network. In this book we suggested looking at
two key centrality measurements: Degree centrality and betweenness. Degree
centrality reflects the number of ties a word has with other words. It can indicate
how much a word is popular in general. Stop-words (for example, pronouns such
as “I”) may be frequently used in various contexts and thus be linked to many
different words. Once we remove stop-words, the words with the highest degree
are often also those that tell us what the text is mainly about. Yet more than
degree, the various studies presented in this book looked at another useful cen-
trality measurement—betweenness. Betweenness reflects the extent that a certain
word is positioned in between any other pair of words as a mediator. The meaning
of this in a semantic network is twofold: It is a word that is located in between
many words within a cluster as well as between clusters. A word located in between
many words within a cluster is often the word that summarizes best the main theme
around which other relevant words evolve. A word located in between different
clusters often also conveys a broader meaning and enables researchers to explore
the meta-narrative of the entire corpus.
Thus, for example, when looking at the Japanese and Korean discourse on Twitter
on sexual abuse in Chapter 7, Mizoroki and Kim found the word “book” to have a
significantly higher betweenness. The cluster of words surrounding the word “book”
included words such as “media”, “BBC”, “documentary”, “news article”, and “TV
program”. Obviously, all these words have one similar meaning in common—the
media channels in which the story of Shiori Ito-, the journalist who was sexually
abused, was told to the public. Yet the network shows that the word “book” is par-
ticularly central in terms of betweenness, indicating that the book Black Box was per-
haps one of the most important pieces of documentation of the story as it contained all
the details of the event and made the victim’s full name public for the first time. In
fact, the other media channels always referred to the book as a reference. The word
book was not only in between media-related words within the media cluster, but also
in between central words of other clusters, such as “rape”, “damage”, and “Abe” (the
former Japanese Prime Minister). It is therefore through the overarching link between
words with the highest betweenness that researchers can identify and extract the meta-
narrative of Twitter users in Japan, who mainly referred to Ito-’s media appearance, the
details of the lawsuit, and its connections to the corruption of the political elite, rather
than to the more significant social problem of gender inequality in Japan.
Look at the composition of clusters. In all the networks presented in the book,
regardless of the type of text in question, it was possible to divide words into
clusters, where each cluster represents distinctive themes and frames. Although
there are many ways to cluster words in semantic networks, most of the studies
here used the Louvain modularity (Blondel et al., 2008). The advantage of this
method is that it compares the relative density and cohesion of sub-networks and
often identifies clusters of words that tend to appear together more frequently in
the same sentence. After inspecting the central words and their context, researchers
should look carefully at the composition of words in each cluster and see what they
have in common and in what context these words appear together in the text.
 Summary and conclusion 223

When looking at the literature on school improvement in Chapter 9, Addi-

Raccah, Shahrabani, and I identified four main themes based on the cluster analysis
of frequent words in academic papers. This map clearly indicated that the focus of
academic papers is heavily influenced by business and management scholarships.
School leadership and teachers were at the center of attention and students were
perceived as passive in the process. Their outcomes were measured quantitatively,
particularly through their achievements in math tests. This investigation helped us
not only to map the literature in the field, but also to reveal its lacuna on which
future studies should focus.
Notice the position of each cluster in relation to other clusters. One of the great advantages
of semantic networks compared to many traditional qualitative and quantitative
analyses is that the clusters identified in the map are positioned in some proximity to
one another. Certain words that appear in between them (see the first point in this
section) can help researchers to understand the relationship between clusters and
build the meta-narrative of the entire text in question.
In studying the role of WhatsApp in the family Noa Hatzir (Chapter 8) used
semantic network analysis to identify themes in interview transcripts. She found
that the clusters containing words that referred to the time aspect (such as “avail-
able” or “convenient”) are located in between two other large clusters—one
reflecting the need for interpersonal connection (with words such as “relationship”
or “depend”), on the one hand, and another reflecting the various platforms (with
words such as “facebook”, “email”, “phone”, and “face-to-face”), on the other.
Hence, the three themes are connected in one meta-narrative in which time con-
straints are the most significant criteria in choosing between different platforms
when maintaining relationships between family members.
Search for the most frequent path between words. So far, we mentioned two ways to
employ semantic networks to locate the meta-narrative: Identifying the words with
the highest betweenness and looking at the relative position of clusters. Another
way, which is certainly related to the other two, is to search for the most frequent
links between words. When using means of visualization, such as highlighting the
width of the links based on their co-occurrences or strength, one may further
notice an overall path that connects the central words together. This path is a
useful method to locate the meta-narrative that can summarize well the entire text.
Locate and study unique and unexpected words in the network. At the beginning of
this section we suggested looking at central words in their network environment,
or in other words, their cluster. Some words would appear in a reasonable place,
surrounded by words with similar meaning, while others would appear in more
surprising and unexpected places. Thus, for example, in the network analysis of
political speeches presented in Chapter 5, together with Maya Hadar and Regula
Miesch we identified the centrality of the words “believe” and “Begin” in the
cluster surrounding the word “I”. The analysis of these words in their context
helped us to offer a much richer and deeper contextualization of Netanyahu’s dis-
course on Holocaust-related speeches and his continuous attempt to associate
himself with the Revisionist ideology of his predecessor, Menahem Begin, that
224 E. Segev

aspires to territorial maximalism on the one hand and his attempt to project an
image of peace negotiator on the other.

The boundaries of semantic networks

Similar to other methods of content analysis, semantic networks are limited to the
content in question. They can offer a window to look at what people write about, but
not always also what they think, believe, or feel. This is certainly the case when
studying news content or political speeches, but also to some extent social media
content. It is important to consider that each type of text is bound to specific social
conventions. Twitter users are limited in the length of their posts and are expected to
comply with a certain style of writing. This is true not only for microblogging services
but more generally to all user-generated content. The power of semantic network
analysis is to map the topics and reveal the possible biases in the text, but not always
beyond. To better understand the views, attitudes, and opinions researchers can
combine semantic network analysis with questionnaires or interviews, which them-
selves can be transcribed and analyzed using semantic networks.
As was indicated in the book, semantic network analysis is powerful in providing
the macro-level picture of the text and its structure. In some of the chapters,
researchers used the descriptive map of the semantic network and the central keywords
they identified to zoom in and inspect the content in a more qualitative approach. It is
very much encouraged to combine semantic network analysis with qualitative content
analysis that brings out some deeper layers of meanings or even with quantitative
codebooks that offer a more detailed view of the content at the meso-level.
Hence, semantic network analysis can only provide a partial picture of what the
text is about. It is a very useful approach as the first stage to map the content. Based
on the categories that emerged (bottom-up) from the network and the categories
that derive from the theoretical framework (top-down) researchers can further
employ other methods, such as quantitative content analysis using a codebook or
qualitative discourse analysis of specific topics or keywords in the text that were
found to be central and interesting for further investigation.
Finally, it was earlier argued that semantic network analysis is particularly pow-
erful when looking at larger texts, containing at least a few thousand words. Still,
when it comes to Big Data analysis, and the size of the text scales up to several
gigabytes of data, the computation resources for semantic network analysis could be
quite demanding. When the available resources and skills are limited, one might
select a random sample of the text, which very often yields similar results. Using
random samples researchers might not be able to locate marginal topics and
exceptional content but could still derive the prevalent themes and their biases.

Future directions in the field of semantic network analysis

The field of semantic network analysis is still relatively young but rapidly devel-
oping, thanks to the growing number of software and online tools available. In this
 Summary and conclusion 225

book we presented the basic principles for you to conduct semantic network ana-
lysis and to use this map to navigate in the text and produce meaningful analyses.
There are numerous directions to develop this method further, including looking
not only at the visualization of networks and their structure based on the centrality
of words and their clusters, but also based on the various measurements that the
network analysis offers.
Distribution of word frequencies and centrality. One of the first indications for the
structure and content is the variety of words in the chosen corpus. A great variety
of words may reflect richer vocabulary and higher language proficiency, but more
importantly, it could also reflect a greater variety of topics and frames in the text.
The distribution of word centrality is even more informative, as it reflects to what
extent words are connected to each other, and how the various topics are formed
to tell one coherent story or parallel stories. In very ideological texts, it is reason-
able to expect a high hierarchy in terms of the frequency distribution and centrality
of words. Most of the words would link to the main central idea, whether it is to
the word “god” in the Quran or a few central concepts such as “class”, “society”,
“proletariat”, and “bourgeois” in the Communist Manifesto (Paranyushkin, 2018).
The different distribution of word frequency and centrality and the way it reflects
the different types of text should be explored more systematically in future studies.
In Chapter 10, Nico Pfiffner made such an attempt to understand the structure
of the text by examining several measurements, including the network density, its
average shortest path length, and its average clustering coefficient. He compared
the scope of several academic journals in the field of communication by looking at
author assigned keywords in academic papers. In many ways, those different mea-
surements tell a similar story about the level of topic coherence in each journal. In
particular, the density of the keyword network in academic journals was a good
predictor of their focus and scope. Journals with higher density were also more
specialized (e.g., Political Communication) than journals with lower density (e.g., New
Media & Society). It seems, however, that due to the great similarity between those
academic journals, there was no significant difference among the other measure-
ments. Future studies should continue and explore the nature and meanings of
these measurements in other settings.
Word centrality measurements. In this book we particularly stressed the importance
of betweenness centrality, as it helps to identify words that connect other words in
the same cluster and thus convey the shared context or meaning of each cluster. At
the same time, words with higher betweenness are located in between clusters and
are therefore useful to consider when trying to extract the overall narrative of the
text. Another advantage of betweenness centrality is that it is unequally distributed
among the words, helping researchers to visually identify the important themes
based on a handful of central words. Yet, apart from betweenness, there are plenty
of other centrality measurements. Future studies should continue and investigate
the unique function of each centrality measurement, and its contribution to
unveiling the meaning of the text.
226 E. Segev

Prevailing path as the main narrative. One of the fundamental principles and biggest
advantages behind semantic networks analysis is its ability to extract and highlight the
association between words. The meaning of words lies exactly in their unique com-
binations. When constructing a semantic network based on large texts, the most fre-
quent word occurrences are the backbone or the meta-narrative of the entire discourse
on the topic in question. Thus, for example, when looking at the coverage of Iran in
Israeli news (Chapter 2), the very prominent link between the words: “sanctions”—
“nuclear”—“weapon”—“peace”—“agreement” summarizes extremely well the
entire idea behind more than 20,000 news items. Similarly, when looking at the social
media discourse around Brexit (Chapter 6), the prominent links between “people”—
“happy”—“#brexitcelebration”—“freedom” on one side of the network, and
between “EU”—“London”—“Scotland”—“Irish”—“sad”—“mistake” on the other
side, brilliantly reflects the polarized discourse in thousands of tweets. In many exam-
ples along the book the prevailing path between central words in the network helps us
to extract and even crystallize the main narrative. Future studies should continue to
explore this technique, which could be particularly useful, for example, in artificial
intelligence applications for an automatic summary of large texts.

Epilogue
Communication is the fabric of society and stories are the essence of communica-
tion. We tell stories to one another by putting words together. The meaning of
words is therefore the meaning of our social being. As social scientists we are
interested not only in one particular story, but in unveiling the patterns behind
many similar stories. We could easily read and make sense of one or a few news
items, but it would be much less feasible to read thousands of them or more. On
the other end, it would be difficult to get the hidden story behind one tweet.
Information nowadays becomes shorter with microblogging and instant messaging
services, and its volume gets much bigger (Segev, 2019). It is therefore much more
challenging for researchers to make sense of the vast amount of information avail-
able. Semantic network analysis as demonstrated in each part of the book is one of
the most powerful and useful methods available to navigate in different types of
texts and unveil their meanings.
But what are all the stories about? Perhaps the DNA of all social science research
could be summarized in one word: Identities. We tell each other stories from the
minute we are born to construct and shape our identities as belonging to one
family, social group, nation, religion, race, and gender, yet not to the other. The
meaning of words, as demonstrated in each chapter of this book, is mostly an
attempt to represent those abstract identities and to add our normative judgement
to them as good and desirable or bad or undesirable.
While some of the chapters focused on abstract identities such as “good” and
“bad” countries and their role and responsibility to the international community,
other chapters paid closer attention to the behavior of different social groups (based
on gender and social role such as students, children, and parents, or even
 Summary and conclusion 227

communication scholars). When constructing and interpreting semantic networks,

it is important to consider that most of the stories we tell and the words we choose
to tell them represent the frames of our identities. The bias of our stories is there-
fore our views regarding what is desirable and included and what is undesirable or
excluded. It is the map for the meaning of our life.

Note
1 As for most figures in this book, the network analysis and visualization software Visone
(Brandes & Wagner, 2004) was used to construct the semantic network for the book
cover.

References
Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of
communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 10,
10008.
Brandes, U. &. Wagner, D. (2004). Visone: Analysis and visualization of social networks. In M.
Jünger & P. Mutzel (Eds.), Graph drawing software (pp. 321–340). New York: Springer-Verlag.
Higuchi, K., (2016). A Two-Step Approach to Quantitative Content Analysis: KH Coder
Tutorial Using Anne of Green Gables (Part I). Ritsumeikan Social Science Review, 52(3), 77–91.
Higuchi, K., (2017). A Two-Step Approach to Quantitative Content Analysis: KH Coder
Tutorial Using Anne of Green Gables (Part II). Ritsumeikan Social Science Review, 53(1),
137–147.
Kitchin, R., & McArdle, G. (2016). What makes Big Data, Big Data? Exploring the ontological
characteristics of 26 datasets. Big Data & Society, 3(1), 1–10. doi:10.1177/2053951716631130.
Paranyushkin, D. (2018). Measuring discourse bias using text network analysis. Towards Data
Science. https://towardsdatascience.com/measuring-discourse-bias-using-text-network-ana
lysis-9f251be5f6f3.
Qin, Wenfeng and Wu, Yanyi (2019). jiebaR: Chinese Text Segmentation. R package
version 0.11.
Segev, E. (2019). Volume and control: The transition from information to power. Journal of Multi-
cultural Discourses, 14(3), 240–257. doi:10.1080/17447143.2019.1662028.
Segev, E. (2020). Textual network analysis: Detecting prevailing themes and biases in interna-
tional news and social media. Sociology Compass, 14(4), 1–14. doi:10.1111/SOC4.12780.
GLOSSARY

API (Application Programming Interface) is the interface that facilitates the com-
munication between applications. In the studies presented in this book API was
used to get access to the Twitter database to collect tweets (Chapters 1, 6, and
7) and to the Crossref database to collect academic papers (Chapter 10).
Betweenness centrality measures the number of the shortest paths in a network
that pass through a node. In other words, it reflects the extent to which a node
is in-between all other pairs of nodes in a network. In semantic networks, this
measurement allows identifying the most central words in each cluster and
words that are located in-between clusters. These words often help to con-
textualize the network maps and understand their meanings (see also Chapter 1
and the concluding chapter for a detailed discussion on betweenness centrality
in semantic networks).
Brexit is a portmanteau of “British exit”, referring to the withdrawal of the UK
from the EU in January 2020. Chapter 6 focuses on #Brexit in Twitter to study
the discourse in social media and compare it to the news coverage of Brexit in
different countries.
City diplomacy refers to the discussion between city officials and the interna-
tional public. Chapter 6 shows that cities, particularly London, are prominent
actors in Twitter in relation to the Brexit discourse. With the growing use of
social media, cities increasingly play an important role in public diplomacy, and
may sometimes depart from the official views and agendas of the government in
their country.
Cluster analysis is a process of grouping data into discrete groups (clusters), in
which the objects of each group share similar traits. In network analysis, clusters
are often determined based on greater connectivity (more links) between certain
nodes. In this book, cluster analysis refers to the process of identifying groups of
 Glossary 229

words that tend to appear together, and thus form a reoccurring theme in the
text.
Clustering coefficient is the degree to which nodes tend to cluster together. In a
sematic network, a high clustering coefficient indicates that words are tightly
connected in clusters, and clusters are distinctive from each other. Chapter 10
uses this measurement to compare the composition of distinctive topics in dif-
ferent communication journals.
Degree centrality measures the number of links a node has with other nodes.
Together with betweenness centrality, it is a useful measurement to identify the
central words in the text and their connection to other words (see also Chapter
1 and the concluding chapter).
Density refers to the actual number of links a network has divided by all possible
links it could have. A dense network contains many links, while a sparse net-
work contains a few links. When preparing semantic networks for analysis there
is a need to find a balance between too dense networks that are difficult to
interpret and read, and too sparse networks that offer only a little information
about the text in question.
Lemmatization is the process of unifying different forms of a word into a single
form. This includes, for example, conjugations (e.g., “believe” and “believing”),
single and plural forms (e.g., “president” and “presidents”), or prepositions that
in some languages are part of the same word. In sematic networks merging these
words is important, as it reduces the complexity of the network and allows to
form more coherent clusters (see also Chapter 1).
Louvain modularity is one of the methods for cluster analysis (see above). In
semantic networks this is a useful way to cluster words together, as it considers
how dense words are connected to each other, that is, how frequently they
appear together. This clustering method often yields a small number of mean-
ingful clusters or themes.
Paradiplomacy refers to the interactions between sub-national entities, such as
cities, regions, or sub-state actors, and the international public. City diplomacy
(see above) is one form of paradiplomacy. Chapter 6 shows that paradiplomacy
is very prominent in the social media discourse on Brexit.
Partisan media refers to media that is systematically influenced and sometimes
affiliated with the government and its parties. Chapter 4 examines to what
extent different television news channels in the US reflect partisanship in their
coverage of social distancing.
Phatic communication refers to communication that does not intend to inform
or share information, but rather to build social bonds. It is often described as
“small talk”, such as greetings or the exchange of general positive gestures.
Chapter 8 shows that WhatsApp groups in the family mainly use phatic com-
munication among its members.
Scientometrics is a research area devoted to the quantitative study of science,
aiming to map and understand the structures of scientific domains. Chapter 10
230 Glossary

makes such an attempt to map the field of communication based on the net-
work of keywords appearing in thousands of academic papers.
Sentiment analysis measures the extent of positivity or negativity expressed in a
text. Chapter 4 shows that semantic networks are particularly useful when
conducting sentiment analysis, as they allow researchers to focus on small seg-
ments of the text and identify the combination of words or the context in
which positive or negative sentiments appear.
Social distancing refers to the physical distance between people as a way to limit
the spread of a contagious disease. During the COVID-19 pandemic, social
distancing has become a common practice, but also a heavily connotated poli-
tical term. Chapter 10 explores the sentiments related to social distancing cov-
erage in different television news channels.
Sparsification is a process in which the density (see above) of a network is
reduced to detect its structure and make it more readable and easier to interpret.
Chapter 1 discusses several ways to sparsify semantic networks, including delet-
ing less frequent links between words and sometimes removing the most central
word that was used as the search term to retrieve the text.
Stemming is the process of reducing words to their word stem form (see also
Lemmatization above). This is a crucial stage in producing readable networks
with more coherent clusters (see also Chapter 1).
Stop-words are commonly used pronouns, prepositions, and conjunctions, such as
“the”, “that”, or “of”. When constructing semantic networks based on frequent
words, stop-words should be removed from the analysis. These words tend to
connect to most other words and usually disturb to detect the network structure
and identify meaningful clusters.
Synchronous/asynchronous communication refers to the real-time exchange
of messages, such as in face-to-face, telephone, or video conversations. Asyn-
chronous communication, on the other hand, involves sending and receiving
messages through mediated means that do not require the real-time presence of
the communicating parties (e.g., sending emails and instant messages or posting
messages in social media). Chapter 8 explores the specific role of WhatsApp as a
channel that allows both synchronous and asynchronous communication, and
therefore has become a popular channel to signal connectedness and express
phatic communication (see above).
Tokenization is the process of word segmentation. In most languages, space is
used to divide between words. In some languages, such as Japanese, tokeniza-
tion is more challenging, and requires the use of specific software. Chapter 7
demonstrates the use of semantic network analysis in Asian languages.
INDEX
#metoo 136–138, 141–142, 144, 146,
148–149, 151–153
ahn 138, 143–144, 149–150, 153
API 17, 28, 30, 119, 143, 200
Begin, M. 97, 103, 108n3, 223
betweenness 6–7, 23, 27–29, 37–45, 47, 49,
61, 63, 84, 101–102, 104, 145, 148, 164,
166–167, 180, 207, 222–223, 225
bible 17, 18, 19, 20, 21, 22, 74
Bild Zeitung 119
brexit 12, 112–131, 217–219, 226
British 12, 35, 55, 112, 115–116, 118, 120,
123–129
China 24, 28–30, 33, 35, 38–39, 53, 56–67
city diplomacy 12, 113–115, 128, 131
cluster analysis 22, 26, 28–29, 37, 61–62, 88,
102, 128, 145, 164, 179–180, 219, 223
clustering coefficient 201–203, 211–212, 225
CNN 12, 73–76, 80–89, 217, 219
co-citation 194–195, 198
country image 12, 113, 116–117, 128
CSV 20
density 22, 179, 201, 202, 211–212, 222, 225
digital diplomacy 28, 115–117
diplomacy 12–13, 24, 28–29, 36–37, 42,
112–117, 119, 121, 127–128, 130–131,
218
emoji 9, 165
ethos 4–5, 102–103, 106, 184
Facebook 16–17, 25, 117, 142, 149, 153,
159, 162, 166, 209, 223
Factiva 17, 20, 119, 131
fake news 1, 24–27, 30, 116
family 13, 20, 103, 159–171, 197, 218–219,
223, 226
FOX News 12, 73–76, 80–89, 217, 219
France 32–33, 38, 60–62, 116–119,
121–122, 126–127, 129–130
Gephi 20
Germany 12, 32–33, 38, 60, 62, 94,
96–100, 104–108, 116–119, 121–123,
126–127, 129–130, 184, 217
Global Times 57–60, 67, 68n2
Haaretz 34, 36–43, 46–68
hashtag 13, 29, 119, 121, 128, 130–131,
140–142, 144, 146, 151–152, 212
holocaust 12, 33, 42, 94–108, 217, 219, 223
Iran 11, 29, 32–33, 35–42, 46–49, 60–62,
64, 97, 102, 106–108, 217, 219, 221, 226
Ireland 112, 118–119, 122–123, 126–127,
129–130, 181
Irish Times 119
Israel 11–13, 32–42, 46–49, 60–62, 94–103,
105–108, 159, 162, 165–166, 170, 217,
219, 226
Japan 13, 18, 49, 56, 59–62, 136–137, 139,
142–149, 151–153, 218–220, 222
KH coder 20, 143, 220
232 Index
Le Figaro 119
Le Monde 119
leadership 58, 94–95, 97, 106, 108, 175,
177, 179–184, 209, 223
lemmatization 18, 163, 201, 212, 220
logos 4, 102–104, 106
London 12, 96, 112, 115, 121, 123–124,
126–129, 217, 226
Louvain 22, 26, 28–29, 37–45, 61, 63, 85,
101–102, 104, 120, 125, 144–145, 164,
166, 180, 219, 222
mediated communication 159–161
memory 3, 12, 94–96, 98–100,
105–107, 145
Merkel, A. 12, 95, 100–101, 103–107, 120
MSNBC 12, 73–76, 80–82, 88–89, 217
national identity 12, 95, 107, 112, 116, 124,
128, 130
Netanyahu, B. 12, 33, 37, 42, 94, 97,
101–104, 106–108, 223
New York Times 24, 27, 30, 57–58, 60,
66, 68n2
Nexis Uni 17, 20, 24, 89, 131
NodeXL 20, 79, 85, 90n3
North Korea 12, 35, 38, 53, 55–57, 59–60,
62, 65–67, 217
nuclear 11, 32–38, 40–43, 46–48, 55, 62,
64–65, 97, 102, 106, 108n3, 108n4, 162,
217, 219, 221, 226
paradiplomacy 13, 112–114, 121, 218
partisan 53, 65, 72–76, 86–87
path length 201–202, 225
pathos 4–5, 102–104, 106
People’s Daily 57–58, 60, 67, 68n2
phatic communication 4, 13, 159–162,
170, 218
political speeches 2, 11–12, 16, 95, 101,
106–107, 216–217, 223–224
populism 113, 121, 126, 128, 130
provocation 10–12, 53–68, 217, 219, 221
public diplomacy 24, 29, 112–114, 119,
128, 130–131, 218
Putin, V. 38–39
Python 20, 23, 89, 143, 200, 202, 212n2
qualitative 2–3, 7–8, 11, 17, 23, 30, 63, 88,
108, 118, 142, 163, 166, 171, 176, 179,
186, 196, 219–221, 223–224
quantitative 2–3, 7, 11, 30, 88, 118, 142,
185, 194, 196–197, 201, 223–224
R 12, 20, 23, 27, 79–80, 144
Russia 32–33, 35, 38–39, 56, 59–65
school improvement 13, 175–186, 218, 223
school effectiveness 175–178, 180–181
scientometrics 194–195
sentiment analysis 76–80, 88–90
Seo, J. 138, 143–144, 148–150, 152–153,
154n3
sexual abuse 136, 138, 151–152, 219, 222
social distancing 12, 72, 74–76, 80–82,
85–90, 217
South Korea 12–13, 53, 56–57, 59–62, 65,
136, 142, 151, 218
sparsification 20–21, 25–26, 221
stemming 18, 80, 163, 220
stop-words 18, 20, 24, 36, 58, 63, 80, 101,
120, 131, 163, 179, 201, 222
Süddeutsche Zeitung 119
synchronous communication 160,
168–171, 218
teachers 175–178, 180–186, 218, 223
tokenization 18, 63
transcripts 3, 7, 10–14, 16–17, 30, 80, 89,
160, 166–167, 170–171, 201, 216–218,
220, 223
Trump, D. 24–25, 29, 33, 38, 64–65, 75,
86–87, 114, 126
Twitter 10, 13, 16–17, 24–25, 28–30, 64,
115–117, 119–123, 126–128, 130–131,
136, 141–143, 148–149, 151–153, 161,
217–219, 221–222, 224
United Kingdom 12, 33, 50, 62, 72,
112–113, 116–119, 121, 123, 125–126,
128–130
United States 10–12, 19, 28–29, 32–33, 35,
38, 47, 53, 55–59, 61–68, 72–74, 120,
124, 142, 193, 217
Visone 20, 25, 29, 38–41, 61–64,
101–102, 104, 120, 144–145, 164,
180, 182, 227n1
Washington Post 24, 27, 30, 57–58, 60, 66,
68n2
WhatsApp 13, 159–171, 218–219,
221, 223
Wikipedia 16
Yamaguchi, N. 137, 143–148, 153
Yedioth Ahronoth 36–37, 39, 41–42, 46–48