Social Media Marketing

Analytics
Session 12
Dr. Mohit Malhan
80% of Data is Unstructured
• Database notes
• Call center transcripts
• Other CRM
• Email
• Open-ended survey
responses
• Web pages
• News Groups
• Documents themselves
• Competitive information
This fact means that decision-makers
often rely on only 20% data. • Reviews, tweets, comments
• Photos, Videos, info graphics.
2
What is Text Analytics?

• Text analytics, also known as text mining, is a

technique to extract, analyze, and interpret hidden
business insights from (unstructured) textual
elements of web and social media content.

3
Types of Text
• Dynamic Text
• Static Text

• Both dynamic and static text are subject to analytics.

4
Types of Text
• Dynamic Text
• Dynamic text is a real-time user-generated text or
statement to expresses an opinion about content or
information posted over social media.

• Dynamic text is usually smaller in length (e.g., a couple of

sentences), diverse in nature, and is updated or deleted
more frequently.
Examples:
• Tweet
• Comments
• Discussion
• Conversation
• Reviews
Copyright © 2018 Gohar F. Khan 5
Types of Text
• Static Text
• Static text is usually large in length (e.g., several
paragraphs) and is generated, updated, or deleted less
frequently.

Examples:
• wiki content
• a blog page
• Word documents
• Corporate reports
• electronic mail (e-
mail), and
• news transcripts.
Copyright © 2018 Gohar F. Khan 6
Deployment Models

• On-premise model
• It is comparatively expensive option, but provides extra
security and control

• Cloud based model

• Attractive to small and medium scale businesses providing
them a cost effectiveness, scalability, and lower risks
solution.

7
Key players

• IBM (Watson Analytics for SM; could based); IBM SPSS

Text Analytics for Surveys (on premise)

• Lexalytics (Semantria) (both)

• Microsoft (Text analytics APIs) (cloud)

• SAS Text (on premise)

• SAP (HANA platform) (both)

8
Activity

• List all possible applications of text analytics (you may

google it).

9
Applications for Text Analysis
• Social media data
• Surveys
• ‘Reading’ email
• Call centre data
• Abstracts
• Document management
• Corporate history
• Scientific publications
• Thematic understanding of website
• Database notes

10
Purpose of Text Analytics
Sentiment
Analysis

Concept Perpose of Text Trends

Mining Analytics Mining

Intention
Mining

11
Purpose of Text Analytics
• Sentiment Analysis
• Sentiment analysis analyzes and categorizes social
media text (mostly dynamic text) as being positive,
negative, or neutral.

12
Purpose of Text Analytics

• Intention Mining
• Intention or intent mining aims to discover users’ intention
(such as buy, sell, recommend, quit, desire, or wish) from
media text.

• Can be used to find new potential customers who

intend to buy a product (or services), and
• service existing customers who have trouble with a
product.

13
Purpose of Text Analytics
• Trends Mining
• Trends mining, also known as predictive analytics,
is used to predict future events.

• Google flue trends (closed now)

• Google correlate (inverse of flue trends)
• Google Trends

• Vast amount of social media data (e.g., comments and

tweets) can be mined to identify patterns and trends
for new product development. 14
Purpose of Text Analytics

• Concept Mining
• Concept mining aims to extract ideas and concepts from
documents.

• Concept mining is useful in extracting ideas from large

amounts of static social media text, such as wiki content, a
web page, Word documents, and news transcripts.

• Concept mining can be employed to classify, cluster, and

rank ideas.

15
Text analytics mechanism

Concept Maps

Text Attract
Attitudes
Clustering
Grow
Surveys Categoriza-
tion

Text NLP Retain

Actions Concepts
Web Trending
Channel Outcomes
Information
Extraction

Business UI
Operational Fraud
Systems
Attributes
Prediction
Customer Data Expert UI Business
Data Collection User

Copyright 2003-4, SPSS Inc. 16

Natural language processing

• NLP is a field of computer science, artificial

intelligence and computational linguistics
concerned extract meaning from text in a way that
is very similar to how the human brain
understands language.
Great service, highly recommended (easy example)

I do not dislike cabin cruisers (challenging example)

17
Supervised vs. Unsupervised Learning

Two classes of machine learning techniques: supervised and unsupervised.

• In supervised the algorithm is trained on the type of data

it can expect to analyze.

• Unsupervised machine learning method try to discover

patterns rather than trying to fit the data into the
predefined structure.

18
Steps in Text Analytics
• Clustering • Dynamic text: tweets,
• Classification comments, reviews
• Association analysis • Static Text: Wiki content,
• Predictive analysis blogs, websites, reports.
• Sentiment analysis

Source
Text Mining Identification

Text Parching
Text &
Transformation
Filtering
• Terms count • Stemming
• Frequency count • Parts of speech
• Co-occurence metrics • Named entities
extraction
• Stop words
• Filtering

Figure 2. Steps in text analytics

19
Steps in Text Analytics
Text Mining
• At this step, the text is actually mined to extract the
needed business insights
• Varieties of text mining algorithms/techniques are
applied to the text, such as:
• Clustering
• Association
• Classification
• predictive analysis, and
• sentiment analysis
20
Steps in Text Analytics
• Clustering http://tweettopicexplorer.neoformix.com/#n=NYTimes
• Clustering or cluster analysis groups objects based
on similarity in non-overlapping groups.

Most request words in

tweets co-occurring together.
Size is proportional to
frequency.

21
Steps in Text Analytics
• Association
• Association or association rule mining is a data-mining
technique used to find frequent patterns, correlations,
associations, or causal structures from data sets.

• Wal-Mart Beer- Diaper example

• It can be used, for example, to find that a user who liked a

social media content A and B is 90 percent likely to also like
content C.

22
Steps in Text Analytics

• Classification
• Is used to find similarities in the document and
groups them with predefined labels based on the
themes contained in the document.

• For example, an e-mail can be classified as spam

based on its contents.

23
What you can do with it
What you can do with it
What you can do with it
Social Media Text Analysis Tools
• Lexalytics: Lexalytics (http://www.lexalytics.com/) is a social media text and semantic analysis
tool for social media platforms, including Twitter, Facebook, blogs, etc.

• Discovertext: Discovertext (http://discovertext.com/) is a powerful platform for collecting,

cleaning, and analyzing text and social media data streams.

• Tweet Archivist: Tweet Archivist (https://www.tweetarchivist.com/) is focused on searching,

archiving, analyzing, and visualizing tweets based on a search term or hashtag (#).

• Twitonomy: Twitonomy (https://www.twitonomy.com/) is a Twitter analytics tool for getting

detailed and visual analytics on tweets, retweets, replies, mentions, hashtags, followers, etc.

27
Social Media Text Analysis Tools
• Netlytic: Netlytic (https://netlytic.org) is a cloud-based text and social
network analytics platform for social media text that discovers social
networks from online conversations on social media sites.

• LIWC: Linguistic Inquiry and Word Count (LIWC) is a text analysis tool for
analyzing emotional, cognitive, structural, and process components
present in individuals’ verbal and written speech samples:
http://www.liwc.net/

• Voyant: Voyant (http://voyant-tools.org/) is a web-based text reading and

analysis. With Voyant, a body of text can be read from a file or directly
exported from a website

28
 Text Analytics Issues

• Lack of a business case

• It is Resources Hefty
• Sentiments are Complex
• Data is Contextual
• Issues with international text

29
Thanks