AUTOMATIC TEXT SUMMARIZATION

MTP Report submitted to

Indian Institute of Technology Mandi
for the award of the degree

of

B. Tech

by

SHUBHAM AJMERA

under the guidance of

Dr. ARTI KASHYAP

SCHOOL OF COMPUTING AND ELECTRICAL ENGINEERING

INDIAN INSTITUTE OF TECHNOLOGY MANDI

JUNE 2015
 CERTIFICATE OF APPROVAL

Certified that the thesis entitled AUTOMATIC TEXT SUMMARIZATION, submitted by

SHUBHAM AJMERA , to the Indian Institute of Technology Mandi, for the award of the
degree of B. Tech has been accepted after examination held today.

Date :
Faculty Advisor
Mandi, 175001
 CERTIFICATE

This is to certify that the thesis titled AUTOMATIC TEXT SUMMARIZATION, submit-
ted by SHUBHAM AJMERA, to the Indian Institute of Technolog, Mandi, is a record of
bonafide work under my (our) supervision and is worthy of consideration for the award of
the degree of B. Tech of the Institute.

Date :
Supervisor(s)
Mandi, 175001
 DECLARATION BY THE STUDENT

This is to certify that the thesis titled AUTOMATIC TEXT SUMMARIZATION, submit-
ted by me to the Indian Institute of Technology Mandi for the award of the degree of B.
Tech is a bonafide record of work carried out by me under the supervision of DR. ARTI
KASHYAP. The contents of this MTP, in full or in parts, have not been submitted to any
other Institute or University for the award of any degree or diploma.

Date :
SHUBHAM AJMERA
Mandi, 175001
Acknowledgments

I would like to take this opportunity to thank my guide Dr. Arti Kashyap for supporting me
throughout the course. She whole heartedly supported me at all the stages of the project. I’d
like to express my heartliest gratitude to her.
I would als like to thank Dr. Dileep A. D., who have offered me their invaluable experi-
ence and knowledge to guide me throughout the project and accomplish the project goals.
Further, I am grateful to my team mates, Saurabh Jain and Rohit Shukla, for all the
contributions they have made to this project.
Shubham Ajmera

i
Abstract

With the increasing amount of information, it has become difficult to take out concise infor-
mation. Thus it is necessary to build a system that could present human quality summaries.
Automatic text summarization is a tool that provides summaries of a given document. In
this project three different approaches have been implemented for text summarization. In all
these three summarizers, sentences are represented as a feature vector. In the first approach,
features like position of sentences, vocabulary intersections, resemblance to title, sentence
inclusion of numerical data are used and model is trained using Genetic Algorithm. In the
second approach, apart from the features used in the first approach, structure of the doc-
ument, popularity of content are also used as features. In the third approach, word2vec
algorithm is used to generate summary.
Keywords: summarization, text, natural language processing, nlp, text mining, wikipedia,
google search, feature extraction, wordnet

ii
Contents

Abstract ii

Abbreviations v

List of Symbols vi

List of Tables vii

List of Figures viii

1 Introduction 2
1.1 Objective . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Methods and Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . 3

2 Background and Related Work 5

3 Work Done (Give an appropriate title) 7

3.1 A Naive Summarizer . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
3.1.1 Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
3.2 Wikipedia Summarization . . . . . . . . . . . . . . . . . . . . . . . . . 9
3.2.1 Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
3.2.2 Summarizer Class . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2.3 Score Calculation . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2.4 User Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.2.4.1 Querying page . . . . . . . . . . . . . . . . . . . . . . 16

iii
 3.2.4.2 Summary Page . . . . . . . . . . . . . . . . . . . . . . 17
3.2.4.3 Webpage interface for non-Wikipedia Context . . . . . 18
3.2.5 Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

4 Experimental Results 20

5 Conclusion and Future Work 25

5.1 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
5.2 Future Score . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

References 26

iv
Abbreviations

NN - Neural Network
FV - Feature Vector

v
Symbols

P
- Summation over all values

vi
List of Tables

3.1 Illustration of Word Similarity . . . . . . . . . . . . . . . . . . . . . . . 12

vii
List of Figures

1.1 Classification of summarization tasks. [1] . . . . . . . . . . . . . . . . . 3

3.1 UML Class diagram of Summarizer . . . . . . . . . . . . . . . . . . . . 9

3.2 TF-IDF Illustration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
3.3 Google Search Result Count Illustration . . . . . . . . . . . . . . . . . . 11
3.4 Call flow of summarizer. . . . . . . . . . . . . . . . . . . . . . . . . . . 13
3.5 Structure of JSON object returned after Extraction and Processing . . . . 13
3.6 Google Result Count Example . . . . . . . . . . . . . . . . . . . . . . . 14
3.7 The Database for TF-IDF: Stemmed words are stored with their correspond-
ing occurrence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.8 Web version of querying page. . . . . . . . . . . . . . . . . . . . . . . . 16
3.9 Mobile version of querying page. . . . . . . . . . . . . . . . . . . . . . . 17
3.10 Web version of summary page. . . . . . . . . . . . . . . . . . . . . . . . 18
3.11 Different screenshots of mobile interface. . . . . . . . . . . . . . . . . . 18
3.12 Mobile interface for quering page of non-Wikipedia content . . . . . . . 19

4.1 Comparison between summarizer using GA approach and Wikipedia sum-

marizer with Compression rate of 30% against the hand-written summary. 21
4.2 Comparison between summarizer using GA approach and Wikipedia sum-
marizer with Compression rate of 20% against the hand-written summary. 22
4.3 Comparison between The GA summarizer and the Word2Vec summarizer. 22
4.4 Comparison of summaries generated between Wikipedia summarizer and
Microsoft Word. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

viii
4.5 Comparison of summaries generated between GA summarizer and Microsoft
Word. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

1
Chapter 1

Introduction

Accoring to WordNet(Princeton) summary is defined as “a brief statement that presents

the main points in a concise form”. Automatic Summarization is a process of generating
summaries by a computer program.
Summarization process involves interpretation, transformation and generation. There are
two types of summarization: abstraction-based summarization and extraction-based sum-
marization. In abstract-based summarization, an abstract is created by interpreting the text
contained in the original document and generating summary that express the same in a more
concise way. In extractive-based approach some sentences from the original text are taken
out to form a meaningful summary. In this approach sentences are given scores based on
different feature and sentences with higher rating are selected for summary. It uses various
Natural Language processing approaches for information retrieval.
Sumamrization process can also be classified based upon the number of source documents,
task specific contraints and use of external resources as shown in the Figure 1.1. Sum-
marization is classified as single-document or multi-document based upon the number of
source document. In multi document summaization information overlaps between different
document makes task difficult. Based upon external resources summarization can also be
classified as knowledge-rich or knowledge-poor. Knowledge rich summarizer uses external
source external corpus like Wikipedia, WordNet etc. In query-focussed or query oriented
summarization summary is constructed with information relevant to the query. In update
summarization summarizer makes use of current trends for construction of the summary.

2
The purpose of the update summary is to identify new pieces of information from the docu-
ment [1].

Fig. 1.1: Classification of summarization tasks. [1]

1.1 Objective
The objective of the project is to provide summaries of document using different summariza-
tion techniques and to compare the quality of the summaries generated. Different leaning
models, data modelling techniques are also tested like Neural Networks(NN), Genetic Al-
gorithms(GA).

1.2 Methods and Techniques

In this project we have implemented three different techniques for extractive based text sum-
marization. Mainly the approaches used are different in terms of features used and machine
learning model. In the first approach, we used sentence position, positive keywords in sen-
tence, negative keywords in sentence, sentence centrality, sentence inclusion of name entit,

3
sentence inclusion of numerical data, and sentence relative length [2]. The model is trained
using Genetic Algorithms [2]. In the second approach, we chose to work on documents
which are well structured and in which sentences are less connected. For summarization
using this techniue, we have used Wikipedia articles, as they provide structured content,
and sentences are less connected. In this approach google search, wordnet word similar-
ity [3], tf-idf, and sentence position featues are used. In the third approach, continuous bag
of words and skip-gram architecture are implemented using Word2Vec toolkit [4] and neural
networks are used to train the summarizer.

4
Chapter 2

Background and Related Work

Abdel Fattah, Mohamed et al. [2] has proposed a simple approach for text summarization.
They have considered features like position, length, name entities, numerical data, bushy
paths, vocabulary overlaps etc. to generate summary. In this approach, sentences are mod-
elled as vectors of features. Sentences are marked as correct if they are to be put in summary
while are marked as incorrected if not. While making the final choice of sentences, each sen-
tence is given a value in between 0 and 1 and using a machine learning model, the sentences
are selected using those scores.
Various other papers have also been published which are very useful for a particular
class of document. Kamal Sarkar et al. [5] is built for summarization of medical documents
using machine learning approach. Various features are very comman and specific to medical
documents. It uses the concept of cue phases such that if a sentences contains n cue phases,
it gets n as its score for the feature. It also uses position of cue phases in the document
such that if it appears at the beginning or at the end of the sentence, it get an additional 1
point. Acronyms are also used as a feature and sentences having these gets extra points. In
some papers [6], sentences are also broken down by special cue makers and sentences are
represented as a feature vectors.
Ryang, D Seonggi et al. [7] proposed a method of automatic text summarization with
reinforcement learning. Researches have also been done for summarization of Wikipedia
articles. Hingu , D et al. [8] implemented various method for summaring Wikipedia pages.
In one of their methods, sentences containing citations are given higher weigtage. In the

5
other approach, the frequency of words are adjusted based on the root form of the word. The
words are stemmed with the objective to assign equal weigtage to words with the same root
word.
Edmundson (1969) [9] proposed an approach of extraction-based summarization using
features like position, frequency of words, cue words and the skeleton of an article by man-
ually assigning weight to each of these features. The system was tested using manually
generated summaries of 400 technical documents. The results were good with 44% of sum-
maries generated by it were matching the manual summaries.

6
Chapter 3

Wikipedia Articles Summarization

3.1 A Naive Summarizer

A naive summarizer was implemented in Java and the features used were:

1. f1 = Sentence Position: Sentences appearing in the beginning and at the end of the
document are given higher weightage. [2]

2. f2 = Positive keyword in the sentence: It is the frequency of keyword in the sum-

mary. [2]

n
1 X
S core f 2 (s) = t fi ∗ P(keywordi |keywordi ) (3.1)
Length(s) i=1

Where, s is a sentence, n is the number of keywords in s, tfi is the occurring frequency

of keywordi in s. We divide the value by the sentence length to avoid the bias of its
length.

P(keywordi |s ∈ S )P(s ∈ S )
P(s ∈ S |keywordi ) = (3.2)
P(keywordi )

#(sentences in summary and contains keywordi )

P(keywordi |s ∈ S ) = (3.3)
#(sentences in summary)

7
 #(sentences in the training corpus and also in the summary)
P(s ∈ S ) = (3.4)
#(sentences in the training corpus)

#(sentences in the training corpus and contains keywordi )

P(keywordi ) = (3.5)
#(sentence in training corpus)

3. f3 = Negative keyword in the sentence: The keywords that are unlikely to appear in
summary. [2]

n
1 X
S core f 3 (s) = t fi ∗ P(s < S |keywordi ) (3.6)
Length(s) i=1

4. f4 = Sentence Centrality: It is the overlap in vocabulary of the sentence and rest of

the document. It demonstrate similarity of the sentence with the document. [2]

Keywords in s ∩ Keywords in other sentences

S core f 4 (s) = (3.7)
Keywords in s ∪ Keywords in other sentences

5. f5 =Sentence Resemblance to the title: It is the overlap in vocabulary between the

sentence and the title of the docuement. [2]

Keywords in s ∩ Keywords in title

S core f 5 (s) = (3.8)
Keywords in s ∪ Keywords in title

3.1.1 Architecture

The project is built on Java. The process of summarization begins with processing of input
document which is broken down into sentences and subsequently into words. Summarizer
maintains a list of sentences of the document and each sentence is responsible to store the
words contained in it. A UML class diagram of the same is shown in Figure 3.1.

8
 Fig. 3.1: UML Class diagram of Summarizer

The summarizer calculates score of each feature for every sentence. It uses Apache
OpenNLP for name finder(dataset - en-ner-person.bin), location finder(dataset - en-ner-
location.bin), and data/time finder(dataset - en-ner-person.bin, data/en-ner-time.bin). After
getting scores, it adds them up by giving equal weightage to each feature value using equa-
tion 3.9 and returns final score and based upon which the summary is generated.

i
X
S core(s) = S core fi (3.9)

3.2 Wikipedia Summarization

“Wikipedia is a free-access, free content Internet encyclopedia, supported and hosted by the
non-profit Wikimedia Foundation” [10].
Wikipedia Summarizer provides summaries for given Wikipedia pages. The project is
built using Django Web Framework, and uses features like Google Search result count, word

9
similarity using Wordnet, sentence position etc. A custom summary option is also provided
such that user can give keywords which he wants/doesn’t want in summary.

3.2.1 Methodology

The summarizer fetches pages from Wikipedia and filters out tags, references and other
meta content from it. The processed input page is then broken down section. Each section
maintains a separate list of sentences contained in it. After processing and storing of page
content, summarizer calculates feature values for each sentences. List of features used are
as follow:

1. Sentence Position: It is a traditional method for providing score for the sentences [2].
Each sentence in a section is given a score based on its relative position in its section.
Sentences appearing in the beginning of the section are given higher weightage as
shown in equation 3.10.

1
S core(s) = (3.10)
position in the section

2. TF-IDF: TF-IDF for a word in a sentence is inversely proportional to the number of

documents which also contain that word. Words with high TF-IDF numbers imply a
strong relationship with the sentence they appear in, suggesting that if that word were
to appear in a sentence, it could be interest to the user [5]. Data frequency values are
constructed using 1000 randomly selected Wikipedia pages. Higher the frequency of
the word in corpus, lower will be its value. The score of sentence is calculated using
equation 3.11

P 1
wordi ∈W occurance o f wordi in the corpus
S core(s) = (3.11)
No o f words contained in the sentence
where W is the list of words contained in the sentence. An illustration is shown in
figure 3.2.

10
 Fig. 3.2: TF-IDF Illustration

3. Google Search Result Count: This is one of the most important feature in which
importance of section is computed by comparing Google Search result count. A search
is performed for each section using query formed by concatenating section heading
and title of the document. Score are given relative to the section which has most
number of counts as shown in figure 3.3.

Fig. 3.3: Google Search Result Count Illustration

4. Word Similarity: This feature provides customizability to the summarizer, in which a

11
 user can select keywords that he wants/doesn’t want in the summary. The summarizer
then compares similarity of that word with every word in the sentence using Wordnet
dataset. If a sentence contains similar words then depending upon the preference of
user it scores the sentence. For positive keywords, it increments the score for sentences
containing similary words on the other hand for negative keywords, it decrements the
score for sentences containing similary keywords. Some examples of words similarity
are shown in Table 3.1.

Word1 Word2 Score Explaination

cat tiger 0.5 Both belong to Felidae family.
apple tiger 0.1 Less related
apple orange 0.25 Both are fruits
cat cat 1.0 Same words
cat feline 0.5 Synonyms
history past 0.5 Both referring to time

Table 3.1: Illustration of Word Similarity

3.2.2 Summarizer Class

Wikipedia summarizer is implemented in Python and using Django Web Framework. It is a

multi-threaded tool which performs feature extraction concurrently as shown in figure 3.4.
Detailed implementation of each component is provided below:

12
 Fig. 3.4: Call flow of summarizer.

1. Extractor: It extracts out content from Wikipedia pages using urllib2. Extractor is
provided with the title of the Wikipedia article, using which it forms url to extract out
Wikipedia pages.

url = ’en.wikipedia.org/w/index.php?action=raw&title=’ + title

2. Text Processor: After receiving data from the extractor, it removes various content
including references, tags, tables, and unwanted sections. Then it breaks down the
article into section using regular expressions and returns a JSON object to the sum-
marizer as shown in Figure 3.5

Fig. 3.5: Structure of JSON object returned after Extraction and Processing

3. Image Extractor: It provides image to the front-end. It extracts image from infobox
of the Wikipedia page using BeautifulSoup

13
4. Summarizer: After receiving processed data, it starts evaluating scores of sentences
for different feature. It concurrently calculates sentences’ score for faster processing.
It eventually calculates score of each sentences.

5. Google Search Result Count: It uses selenium web driver and BeautifulSoup for
extracting Google Search result count. Selenium web driver uses mozilla to open
Google result page for the input query. After that the text from the page is processes by
BeautifulSoup and subsequently count is fetched from there using regular expressions.
It takes a pause of two seconds before making a new query.

Fig. 3.6: Google Result Count Example

6. TF-IDF: A database of stemmed words is generated from randomly selected 1000

Wikipedia articles. Using the same database, and for the input Wikipedia article,
words are stemmed and scores are calculated using Equation 3.11 for calculating the
score of the sentence. A glimse of database content is shown in Figure 3.7.

14
Fig. 3.7: The Database for TF-IDF: Stemmed words are stored with their corresponding
occurrence

7. Positive Keywords/Negative Keywords: This is an important feature of the summa-

rization. It is used for generating customizable summaries based on user requirements.
For input keywords, similarity is calculated using Wordnet data. For each sentences,
a word is selected which has the maximum similarity with the input keywords and
then score of the sentence is calculated using the selected word.

If the keyword is positive then,

S core(s) = 1 + similarity value (3.12)

If the keyword is negative then,

S core(s) = 1 − similarity value (3.13)

3.2.3 Score Calculation

Final scores of each sentences are calculated by multiplying the value of each feature for a
sentence as shown in equation 3.14

15
 Y
S core(s) = fi (3.14)
fi ∈F

where F is the feature vector for the sentence s.

3.2.4 User Interface

3.2.4.1 Querying page

In the webpage, three boxes have been provided for the Wikipedia page for which summary
has to be generated, important keywords and negative keywords as shown in Figure 3.8.

Fig. 3.8: Web version of querying page.

A mobile interface is also developed for the same as shown in Figure 3.9

16
 Fig. 3.9: Mobile version of querying page.

3.2.4.2 Summary Page

This page is built using Bootstrap and Javascript. This page gets GET content from the
querying page using which it fetches relevant summary, image and title of the document.
It also has a slider using which a user can select the percentage shrinkage according to his
requirement as shown figure 3.10 and figure 3.11

17
 Fig. 3.10: Web version of summary page.

Fig. 3.11: Different screenshots of mobile interface.

3.2.4.3 Webpage interface for non-Wikipedia Context

An interface for non-wikipedia context is also developed in which data can be put manually
which includes document’s title, sections’ heading and its content. According to the need,

18
sections can be added dynamically.

Fig. 3.12: Mobile interface for quering page of non-Wikipedia content

3.2.5 Features

1. Variable length summaries can be generated.

2. Custom summary option is provided to user to remove unwanted topics.

3. Multithreaded feature extraction.

4. Stores summaries of articles so to provide quick results in future.

19
Chapter 4

Experimental Results

For quality comparisons of summaries generated by summarizers, we have manually created

summaries of various Wikipedia articles and anime data. We have generated summaries of
10 Wikipedia articles and 10 anime article. The comparison is done between the Wikipedia
Articles and Genetic Algorithm summarizer using the manually generated summaries of
Wikipedia articles while in between GAS and Word2Vec summarizer using summaries of
animes articles. Also, the comparisons are performed between the summarizers and MS-
Word summarizer. Results and explainations of these experiments are as follows:

1. GA summarizer VS Wikipedia summarizer with CR 30%

In this comparison, the output of the generated summary is compared with the hand-
written summaries of Wikipedia articles using the equation 4.1. Higher the intersec-
tion of sentences between the summaries, higher will be the scores.

20
Fig. 4.1: Comparison between summarizer using GA approach and Wikipedia summarizer
with Compression rate of 30% against the hand-written summary.

By observing the figure 4.1, we have found out that the average rate of intersection
of GA is coming out in between 0.4 and 0.5 with high deviation while the Wikipedia
summaries have shown more consistency with values ranging from 0.7 to 0.8.

#(sentences in s1 ∩ sentences in s2)

S core(s1, s2) = (4.1)
#(sentences in s2)

2. GA summarizer VS Wikipedia summarizer with CR 20%

Again by observing the graph in the figure 4.2, it can be said that Wikipedia summa-
rizer is giving higher percentage of matching as compared to the summarizer using
GA approach. However, with comparison to the graph in figure 4.1, we can see that
percentage similarity between the hand-written summaries and the summaries gener-
ated by two summarizers has decreased.

21
Fig. 4.2: Comparison between summarizer using GA approach and Wikipedia summarizer
with Compression rate of 20% against the hand-written summary.

3. Summarizer using GA approach VS Word2Vec Summarizer

For the comparison of these two approaches, 10 hand-written summaries of anime ar-
ticles are used and scores are calcuated using the equation 4.1. It can be observed from
the graph that there is no much difference between the outcome of the two summarizer
and the range is varying from 0.35 to 0.6.

Fig. 4.3: Comparison between The GA summarizer and the Word2Vec summarizer.

22
 4. Wikipedia Summarizer vs Microsoft Word Summarizer
For the comparison of the summarizers, the Wikipedia articles are fetched to Microsoft
Word and summaries are compared with the equation 4.1. In Contrast to the compar-
ison with the hand-written summaries where the Wikipedia summarizer had shown
very less deviation with a high precision value, has shown a wide variation between
the range 0.38 to 0.6 as shown in graph in figure 4.4.

Fig. 4.4: Comparison of summaries generated between Wikipedia summarizer and Mi-
crosoft Word.

5. GA Summarizer vs Microsoft Word Summarizer

The summarizer has shown similarity in results with the Wikipedia summarizer as
shown in graph in figure 4.5. There is a moderate deviation with a low percentage of
similarity between 0.45 to 0.55.

23
Fig. 4.5: Comparison of summaries generated between GA summarizer and Microsoft
Word.

24
Chapter 5

Conclusion and Future Work

In this project, we have implemented various techniques for extraction-based text summa-
rization and experimented their performance analysis using various datasets.

5.1 Conclusion
1. The summarizer built using genetic algorithms has shown satifactory results for wide
variety of documents.

2. The Wikipedia summarizer, though not quite generic, has shown excellent results for
generating summaries for Wikipedia pages.

3. The Word2Vec summarizer is very much dependent on the training data but shows
good result as compared to other two for the data which is closely knitted.

5.2 Future Score

1. Building a single system which can automatically identify the type of input text and
use the best method out of these three technique to generate summary.

2. Improving accuracy of features like name entity identifier, location finder etc.

3. Extending the domains for Word2Vec summarizer by training it using various other
datasets in diverse domains.

25
References

[1] G. Sizov, “Extraction-based automatic summarization: Theoretical and empirical in-

vestigation of summarization techniques,” 2010.

[2] M. A. Fattah and F. Ren, “Automatic text summarization,” World Academy of Science,
Engineering and Technology, vol. 37, p. 2008, 2008.

[3] “About wordnet - wordnet - about wordnet,” https://wordnet.princeton.edu/, accessed:

2015-05-15.

[4] “word2vec - tool for computing continuous distributed representations of words. -

google project hosting,” https://code.google.com/p/word2vec/, accessed: 2015-05-15.

[5] J. Ramos, “Using tf-idf to determine word relevance in document queries,” in Proceed-
ings of the first instructional conference on machine learning, 2003.

[6] W. T. Chuang and J. Yang, “Extracting sentence segments for text summarization:
a machine learning approach,” in Proceedings of the 23rd annual international ACM
SIGIR conference on Research and development in information retrieval. ACM, 2000,
pp. 152–159.

[7] S. Ryang and T. Abekawa, “Framework of automatic text summarization using rein-
forcement learning,” in Proceedings of the 2012 Joint Conference on Empirical Meth-
ods in Natural Language Processing and Computational Natural Language Learning.
Association for Computational Linguistics, 2012, pp. 256–265.

[8] D. Hingu, D. Shah, and S. S. Udmale, “Automatic text summarization of wikipedia

articles,” in Communication, Information & Computing Technology (ICCICT), 2015
International Conference on. IEEE, 2015, pp. 1–4.

[9] H. P. Edmundson, “New methods in automatic extracting,” Journal of the ACM

(JACM), vol. 16, no. 2, pp. 264–285, 1969.

26
[10] “Wikipedia - wikipedia, the free encyclopedia,” http://en.wikipedia.org/wiki/
Wikipedia, accessed: 2015-05-15.

27
Curriculum Vitae

Name: Shubham Ajmera

Date of birth: 09 October, 1992

Education qualifications:

• [2011 - 2015] Bachelor of Technology (B. Tech),

Computer Science and Engineering,
Indian Institute of Technology Mandi
Mandi, Himachal Pradesh, India.

Permanent address:
‘Keluka Bhawan’,
3/219, Near CTS Bus Stand, Sanganer
Jaipur, Rajasthan, India, 302029.
Ph: +918894849404

28