You are on page 1of 6

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/224129737

Direct automatic generation of mind maps from text with M2Gen

Conference Paper · October 2009


DOI: 10.1109/TIC-STH.2009.5444360 · Source: IEEE Xplore

CITATIONS READS

9 3,740

6 authors, including:

Mohammad Abdeen Mustapha C.E. Yagoub


University of Ottawa University of Ottawa
31 PUBLICATIONS   176 CITATIONS    345 PUBLICATIONS   2,577 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

RFID Technology View project

Employing Takaful Islamic Banking through State of the Art Blockchain: A Case Study of Saudi Arabia View project

All content following this page was uploaded by Mohammad Abdeen on 06 February 2017.

The user has requested enhancement of the downloaded file.


TIC-STH 2009

Direct Automatic Generation of Mind Maps from text


with M2Gen

M. Abdeen, R. El-Sahan, A. Ismaeil, S. El-Harouny, M. C.E. Yagoub


M. Shalaby The school of Information Technology and Engineering
The Faculty of Computers and Information Sciences University of Ottawa
Ain-Shams University Ottawa, Ontario,
Cairo, Egypt myagoub@site.uottawa.ca
mabdeen@alumni.uottawa.ca

Abstract—A mind map is a diagram used to represent words, central keyword. As an example, 0 depicts a mind map of
ideas, or other items linked to and arranged around a central Google tools.
keyword or idea. Mind maps are used to generate, visualize,
structure, and classify ideas, and as an aid in organization, study,
project management, problem solving, decision making, and
writing. It has been long used in brainstorming and as an
effective educational tool.

There are numerous tools in the market, either as freeware or as


proprietary software, that help users generate mind maps..
However, these tools are more of mind map “editing” tools to
help users project their ideas from their minds into the tool
mapping space. These tools also provide a comprehensive library
of images that suits the most popular mind map types. The tools
act as the media into which users projects the maps that has
already more-or-less matured in their minds.
Figure 1 Google tools mind map
In this work, we present a software tool that automatically
generates mind maps directly from text. This tool provides a Manually constructing mind maps requires thorough
prospect to transform many literatures automatically into mind reading and good understanding the text which takes much
maps. One significant application of this tool is education. Many time and effort. In addition to that not all people are creative
students finds it easier to follow and remember information enough to draw elegant and expressive mind maps. Therefore,
presented in the mid map form rather than pure text. automatically generating mind maps saves much time and
effort and serves better and quicker various applications.
Keywords- text processing; semantic analysis; web mining.
Mind mapping applications are numerous. Organizing,
meetings, planning, note taking, presentation, and above all, in
I. INTRODUCTION education [2].
Mind mapping is a popular brainstorming tool and thinking There are numerous tools in the market, either as freeware
technique of visually arranging ideas and their or as proprietary software, that help users generate mind maps.
interconnections. It is a way of representing associated
Wisemapping and Mindomo [3] are examples of the freeware
thoughts with symbols rather than with extraneous words. The
human mind forms associations almost instantaneously, and and Buzan’s iMindMap and Inspiration [4] are examples of the
"mapping" allows capturing these ideas quicker than proprietary ones.
expressing them using only words or phrases. These softwares help the user in drawing the mind map and
Originated in the late 1960s by Tony Buzan, [1] mind have some ready designs and diagrams which can be used. But
mapping harnesses the full range of cortical skills - word, the user must read, understand the text well and come up with a
image, number, logic, rhythm, color and spatial awareness - in design for the mind map himself.
a single, uniquely powerful manner. In so doing, mind mapping Automatically generating mind maps out of pure text
gives the freedom to roam the infinite expanses of your brain. It requires many stages of text processing. In the following
is now used by millions of people around the world. sections, we provide details of the main modules of the tool and
A mind map is a diagram used to represent words, ideas, the stages used to produce the final mind map.
tasks, or other items linked to and arranged radially around a

978-1-4244-3878-5/09/$25.00 ©2009 IEEE 95


II. ANALYSIS AND DESIGN A precise method for quantifying how similar two word
The M2Gen tool consists of five modules: senses are is called a measure of semantic relatedness. The
Text Meaning Representation is concerned with putting the text
• Morphological Analyzer in a form which best represents its meaning. Finally, the mind
• Parser map conversion module obtains candidate pictures by
• Syntax Analyzer performing a Google search with the name of each noun
• Semantic Analyzer (which further consists of three (including adjectives) and each verb in the text and retrieving
the first picture found [5], thereby drawing the mind map.
sub-modules) [9],[10].
Figure 2 above depicts these modules and their interaction.
• Mind map conversion
As an illustration for the design details of the tool modules,
The role of each module is as follows: we show the class diagrams for the mind map conversion
module. This module consists of two sub-modules, the mind
The Morphological Analyzer is concerned by how words map viewing management module and the drawing
are constructed from more basic units called morphemes. It management module. 0 and 0 below show the class diagrams
returns all possible morphemes for each word in the text. for these two modules.
For example: Friendly (Adjective) = friend (noun) + ly
(suffix)

The parser returns all possible parse trees for each sentence in
the text according to the English grammar rules in effect. A
filtering process takes place in which the grammatically correct
parse trees are chosen for each sentence. The Syntax Analyzer
is the module which produces the final correct parse trees of
the input text. Further filtering is done based on a score given
to each parse tree based on its internal structure. The Semantic
Analyzer receives correct parse trees of the text, selects correct
meaning for each word and produces a new Text Meaning
Representation (TMR) or uses it to update an existing one. The Figure 3 The class diagram for the Mind Map conversion module
Semantic Analyzer consists of three sub-modules: the
Discourse Analyzer, the Word Sense Disambiguation and the
Text Meaning Representation. The Discourse Analyzer is
concerned with assigning each pronoun to the noun which this
pronoun refers to. The Word Sense Disambiguation is
concerned with assigning the most proper sense for each word
according to the formulation of the sentence. For example, the
word "ball" has several senses, including a round object used in
games, a formal dance, and a pitch in baseball that is not a
strike.

Figure 2 Schematic representation of mind map generation modules

Figure 4 The class diagram for the drawing management module

96
III. IMPLEMENTATION
As mentioned previously, the tool consists of five modules.
In this section we present the design, implementation notes and
testing of the modules. For the morphological analysis module,
we used the Microsoft WordNet 2.1 analyzer [6]. Generating a
mind map requires the knowledge of the verb tense (for verbs)
while for adjectives it is needed to know if it is comparative or
superlative. The following modifications had to be introduced
to the WordNet so that it can be used in mind map generation:
1. For regular verbs and adjectives, the rules return the
removed suffix to get the morph as additional output.
The verb tense is determined accordingly (past for
“ed” and present for “s”).
2. For irregular verbs and adjectives the reverse of their
exceptional lists were built. That list will later be
searched to determine a verb tense.
In the remainder of this section we show in details, and as
an illustration, the design of the syntactic and the semantic
analyzer modules. Figure 5 A detailed class diagram for the Parser class

A. The Syntactic analyzer module


This module consists of three classes. The Rules class, the 3) The Syntax Analyzer class
Parser class, and the Syntax class. This module is responsible for modification, addition,
deletion or selection of parse trees produced by the Parser. If a
1) The Rules class parse tree doesn’t follow the syntax structure of its verbs, the
It is a simple class where it reads the rules from the parse tree is deleted. The WordNet 2.1 verb syntax structure is
“rules.txt” file to be applied later by the parser so that it can used. Each parse tree is given a certain score based on its
build parse trees. structure. Parse trees with maximum score are selected. The
selection is based on grammatical and statistical basis.
2) The Parser class
This class contains the definition of two data structures The following shows an example to test this capability of
namely the parse tree and parse node. It is also responsible for the tool. An input text of “Shakespeare is a great writer in the
constructing all the possible parse trees to the text. history of literature. He was born in 1564. …”. 0 shows the
output that corresponds to the previous text. The output is
The Parser generates all correct parse trees for every input divided into sentences and for each, syntax analysis is
sentence according to grammar rules. It consists of three parts: performed.
the lexicon, the grammar, and the parsing algorithm. For the
Lexicon, the WordNet 2.1 lexical database is used as a lexicon.
It includes a total of 155327 words. WordNet 2.1 lexical
database does not include pronouns, propositions or
conjunctions. Lists of pronouns, propositions and conjunctions
were added in separated files. For the Grammar rules, Context-
free grammar rules were created for the English language in
“rules.txt” file. It consists of 651 CFG rules. Finally for
parsing, a Top-down chart parsing algorithm is
implemented [11]. 0 depicts a class diagram for the parser
class.

Figure 6 The output of the syntax analyzer module

97
B. The Semantic analyzer module
Semantic analysis is the process of relating syntactic
structures, from the levels of phrases, clauses, sentences and
paragraphs to the level of the writing as a whole, to their
language-independent meanings, removing features specific to
particular linguistic and cultural contexts. It has three functions
and each is performed by a separate module.
1) Word Sense Disambiguation (WSD):
This module is concerned with assigning the most proper
sense (meaning) for each word according to the structure of the
sentence.
2) Discourse analysis
This module is basically concerned with assigning each
pronoun to the noun which this pronoun refers to. It includes
the implementation of the RAP algorithm [12] that is Figure 7 The output of the discourse analysis module of the tool
concerned with assigning an antecedent noun to each pronoun.
It depends on:
3) Meaning Representation class
• Selecting the nouns in the context.
The verbs, adverbs, nouns and adjectives in each sentence are
• Eliminating the nouns which do not agree with the extracted along with any relations (case roles) between words.
pronoun in the person, number and gender. This sub-module is totally dependent on the correctly chosen
parse trees from the Syntax Analyzer. A Case Role is an
• Adding salience values to each noun according to certain
argument or typical role that a predicate can take. It appears as
criteria.
a property of an event (verb) in a TMR. For example, in the
• Selecting the noun with the maximum salience as the sentence "Shakespeare was born in 1564," "born" is the event
anaphora (antecedent) of the pronoun. (verb), the verb is passive therefore "Shakespeare" is the
theme case role (the party that was affected by the verb),
We have tested this module with the following text excerpt:
"1564" is the time case role of the event. There are several
“Shakespeare is a great writer in the history of literature. other case roles including: Agent (The party performing the
He was born in 1564 and he had 3 children. He was earning action), Source, Destination, Location, etc.
his living from buying and selling the agricultural products. He
lived in Stratford. He died in 1616” The following Figure is the Text Meaning Representation for
The output is as follows: the Shakespeare excerpt that was used above.

SHAKESPEARE (1.1), HE (2.1), HE (3.1), HE (4.1), HIS


(4.4), HE (5.1), HE (6.1)
The first word of the output is the antecedent noun that the
pronouns refer to .In our example, Shakespeare and (1.1)
means that Shakespeare is the word number 1 in sentence
number 1 (sentence number “dot” word number). “He (2.1)”
means that "He" is the word number 1 in sentence number 2,
which also refers to Shakespeare in the first sentence. The
same thing goes for "He (3.1)" and so on.

Figure 7 below shows the output of the “Discourse Analysis”


module of the tool.

Figure 8 Text Meaning Representation

IV. RESULTS
To demonstrate the tool’s capabilities we have run a simple
test on a text defining Shakespeare, the great writer. The text is
the one listed in the previous section.

98
The following Figure is the final mind map. it is presented as a separate mind map with that certain node as
the central node or main topic. We also have current work in
this area.
In addition to that, we are working on improving the image
production module to be able to synthesize images form a very
small and manageable database other than using Google image
search facility. This has the advantage of allowing the tool to
work offline and to eliminate any network delays during the
search process.

ACKNOWLEDGMENT
The authors would like to thank Mr. M. Hamdy for his help
and support during the course of the work.

REFERENCES

Figure 9 Final mind map of the Shakespeare text [1] “The Mind Map Book: How to Use Radiant Thinking to Maximize Your
Brain's Untapped Potential” Tony and Barry Buzan, E P Dutton 1994
It is to be noted that some of the images assigned to the [2] David A. Watkins & John B. Biggs. “Teaching the Chinese learner:
Psychological and pedagogical perspectives” Hong Kong : Comparative
nodes are not accurately representing them. This is essentially Education Research Centre 2001.
due to the way Google image search is tuned to perform image [3] http://www.wisemapping.com
retrieval. There are two approaches to solving this problem. [4] http://www.imindmap.com/
The first is to choose tighter search criteria for Google that [5] “An API for Google Image Search”, http://www.codeproject.com/
eliminates any ambiguity in the search. The second is to create KB/IP/google_image_search_api.aspx
a complete image database. The latter option is what we [6] George A. Miller, Christiane Fellbaum ,WordNet 2.1 Documentation,
adopted in the improvements we are currently performing on Cognitive Science Laboratory Princeton University
this tool. [7] Bell C., Bowden M., and Trott A. “Implementing flexible learning.
London” Kogan page Limited, 1997.
V. CONCLUSIONS AND FUTURE WORK [8] "Lexical Semantics & Word Sense Disambiguation", CMSC 35100,
May 15, 2003
In this paper, we presented M2Gen, a software tool that [9] Diana McCarthy, Rob Koeling, Julie Weeds, John Carroll, "Using
automatically generates mind maps from input text. To the best Automatically Acquired Predominant Senses for Word Sense
of our knowledge, this is the first time this tool is presented. Disambiguation", Department of Informatics, University of Sussex, 2003
Due to the size of the English language and the variety of rules [10] Mikyoeng Kim, "A Study on the Algorithms for Dialogue
within it, the current version of the tool covers a relatively Segmentation", School of Knowledge Science, Japan Advanced Institute
of Science and Technology (JAIST), March 2000.
small subset of the English language. More work is currently
being done to expand the scope of the tool to include a larger [11] H. Stahl, J. Muller, M. Lang, “An efficient top-down parsing algorithm
for understanding speech by using stochastic syntactic and semantic
subset of the language. On the other hand, the current version models”, in the proceedings of IEEE International Conference on
only supports single layer mind map and does not produce a Acoustics, Speech, and Signal Processing 1996.
hierarchical (multileveled) map (i.e., maps within maps). This [12] Long Qiu, Min-Yen Kan, Tat-Seng Chua, "A Public Reference
is because not all the information extracted from the text is Implementation of the RAP Anaphora Resolution Algorithm", In
displayed in one mind map, but only the most important nouns Proceedings of Language Resources and Evaluation Conference (LREC
2004), Lisbon, Portugal, 2004.
and verbs are. This means that if further information needs to
be obtained about a certain node, if there is more information,

99
View publication stats

You might also like