Query Sensitive Comparative Summarization of Search Results Using Concept Based Segmentation

Computer Science & Engineering: An International Journal (CSEIJ), Vol.1, No.5, December 2011
DOI : 10.5121/cseij.2011.1503 31
QUERY SENSITIVE COMPARATIVE SUMMARIZATION
OF SEARCH RESULTS USING CONCEPT BASED
SEGMENTATION
Chitra.P1
, Baskaran.R2
, Sarukesi.K3
1
Dept. of Information Technology, RMK Engineering College, Tamilnadu, India
2
Dept. of Comp. Sci. & Engg. Anna University, Chennai, Tamilnadu, India
3
Hindustan University, Chennai, Tamilnadu, India
1
pce.it@rmkec.ac.in, 2
baaski@cs.annauniv.edu, 3
profsaru@gmail.com
ABSTRACT
Query sensitive summarization aims at providing the users with the summary of the contents of single or
multiple web pages based on the search query. This paper proposes a novel idea of generating a
comparative summary from a set of URLs from the search result. User selects a set of web page links from
the search result produced by search engine. Comparative summary of these selected web sites is
generated. This method makes use of HTML DOM tree structure of these web pages. HTML documents are
segmented into set of concept blocks. Sentence score of each concept block is computed with respect to the
query and feature keywords. The important sentences from the concept blocks of different web pages are
extracted to compose the comparative summary on the fly. This system reduces the time and effort required
for the user to browse various web sites to compare the information. The comparative summary of the
contents would help the users in quick decision making.
KEYWORD
Comparative Summarization, concept segmentation, query based summary
1. INTRODUCTION
The WWW grows rapidly and caters to a diversified levels and categories of users. Web search
engines helps in locating information content and normally provide thousands of results for a
query. Users still have to spend lot of time to scan through the contents of this result set to locate
the required information. It is not feasible for the user to open each link in the result set to find
out its relevance. The performance improvement of the search engines has become the most
important research area to satisfy the needs and expectations of diversified target users.
A small summary generated from the content of web page would be helpful for the users to get an
instant feel about the content without going through the entire content. People can have a concise
overview in short time. This can greatly enhance the retrieval efficiency. Automatic summary
aims to extract some important sentences from original documents to represent the content of the
article.

32
Automatic summary produced by current search engines contains first few sentences of the web
page or the set of sentences containing the query key words. Using this information, users have to
decide which of the listed documents in the search result will be most likely to satisfy their
information need.
This paper is an extension of our work mentioned in [1], proposes a summarization technique,
which extracts query relevant important sentences from a set of selected web pages to generate a
comparative summary which would be beneficial for the users to make informed decisions.
The remainder of this paper is organized as follows: Section 2 provides the motivating examples
for this work, Section 3 discusses about the related research works that have been done in this
field and Section 4 describes concept based segmentation process guided by the webpage’s DOM
tree structure. In section 5, we present the framework for the selection based comparative
summarization system. Section 6 compares this system with few other systems, Section 7
discusses about experimentation results and performance measures and in section 8, the paper is
concluded with a view on further improvement to this system.
2. MOTIVATING EXAMPLE
People normally collect all related material and information before they make a decision about
some product or service before they go for it. For example, parents might be interested in
collecting details like placement, infrastructure, faculty details, etc related to various Engineering
Colleges, before they go for admission. To accomplish this, the user collects placement and other
required details of various Engineering Colleges using search engines like Google, prepares a
comparative statement manually to find out the best option for admission.
The proposed system generates the comparative summary from the set of URLs selected by user
from the search result based on the specified feature set. The comparative summary contains the
text relevant to placement and training, infrastructure details, result details and fee structures from
the selected URLs. This would definitely be helpful to get instant comparative statement.
Another example could be the comparison between the services offered by various Banks. Set of
Banks can be selected from the list of Bank web sites and comparative summary based on feature
keywords like home loan, term deposit, etc would be helpful for users to make quick decisions
about their investment.
3. RELATED WORKS
Summarization in general can be categorized into two types as extraction based and abstraction
based methods. Extractive summary is created by extracting important sentences from the actual
content, based on some statistical measures like TFxIDF, SimWithFirst[2],etc. Abstractive
summary is created by rewriting sentences on understanding the entire content of the original
article by applying NLP techniques. The later technique is more computationally intensive for
large data-sets.
Concept based automatic summarization directly extracts the sentences, which are related to the
main concept of the original document while the query sensitive summarization extracts
sentences according to user queries, so as to fit the interests of users. In multi-document
summarization the sentences are selected across different documents by considering the concept
and diversity of contents in all documents.

33
Query based summarization system to create a new composed page containing all the query key
words was proposed in[3]. Composed page was created by extracting and stitching together the
relevant pieces from a particular URL in search result and all its linked documents but not other
relevant documents of user’s interest.
Segmented topic blocks from HTML DOM tree were utilized to generate summary in [4] by
applying a statistical method similar to TFxIDF to measure the importance of sentences and
MMR to reduce redundancy. This system focused on the summary of only single document.
SimWithFirst (Similarity With First Sentence) and MEAD (Combination of Centroid, Position,
and Length Features) called CPSL features were used for both single and multi document text
summarization in[5]. Both these techniques show better performance for short document
summarization but not suitable for large ones.
Document Graph structure of sentences was used in[6] for text summarization. Similarity scores
between the query and each sentence in the graph are computed. Document graph construction is
an overhead for the summarization process.
Balanced hierarchical structure[6] was utilized to organize the news documents based on event
topics to generate event based summarization. This method focused on news and event
summarization.
This research work focuses on the novel idea of generating the aggregation of document
summaries. This document summarization makes use of concept based segmentation of DOM
tree structure of web pages. This comparative summary is composed of the query sensitive
important sentences extracted from concept blocks of different web pages which would be helpful
for decision making.
4. CONCEPT BASED SEGMENTATION
In general, web page summarization derives from text summarization techniques, while it is a
great challenge to summarize Web pages automatically and effectively. Because Web pages
differ from traditional text documents in both structure and content. Web pages often have
diverse contents such as bullets, images and links.
Web documents may contain diversified subjects and information content. Normally, the contents
of same subjects will be grouped under the same tag. This system utilizes the Document Object
Model (DOM) to analyze the content of the web page. The leaf nodes of DOM tree contain the
actual content and the parent nodes generally contain higher level topics or section headings.
4.1 Concept Based Segmentation Process
Fig. 1 depicts the concept based segmentation using DOM tree structure. The rectangular nodes
represent the HTML tags or the higher level topics and the circular nodes represent the
information content within the tag. These circular nodes from left to right constitute a coherent
semantic string of the content[4].
The DOM trees of the user selected URLs are processed to generate the summary. Leaf nodes are
considered as micro blocks which are the basic building blocks of the summary. Adjacent micro
blocks of the same parent tag are merged to form the topic blocks.
Each sentence in the topic block is labeled automatically based on the PropBank notations [7][8].
The information about who is doing what to whom clarifies the contribution of each term in a

34
sentence to the meaning of the main topic of that sentence[9].This concept-based mining model
captures the semantic structure of each term within a sentence rather than the frequency of the
term alone. This similarity measure outperforms other similarity measures that are based on term
analysis models[10].
These sentences are labeled by a semantic role labeler that determines the words which contribute
more to the sentence semantics associated with their semantic roles in the sentence.
The semantic role labeler identifies the verb argument structures of each sentence in the topic
block. The number of generated labeled verb argument structures is entirely dependent on the
amount of information in the sentence. The sentence that has many labeled verb argument
structures includes many verbs associated with their arguments.
The words contributing more to the meaning of the sentence will occur more number of times in
the verb argument structure of the sentence. Hence these words will have comparatively higher
frequency. ASSERT software which is a publicly distributed semantic labeling tool, is used for
this purpose. Each word that has a semantic role in the sentence, is called a concept[4][9].
Concepts can be either words or phrases and are totally dependent on the semantic structure of the
sentence. List of concept words and their respective frequency of occurrences for these topic
blocks are identified.
Concept based similarity between the topic blocks are measured using the concept lists to identify
the similar topic blocks. Topic blocks having similarity above the threshold value α(0.6), are
combined to form the concept block. Topic blocks having content about the same concept (for
example placement and training in a college web site) will be similar to each other. Topic blocks
containing information about similar concept word (placement) are merged to form a concept
block (placement block having placement details of all departments in the college).
The concept block formation could be done offline for all web documents in the repository. The
concept block id, conceptual terms, frequency and list of sequence numbers of sentences of each
of these concept blocks are stored in the offline database which would be required for processing
at run time. These concept blocks contain related information content scattered throughout the
Home Dept Admission
html
head body
CSE
IT ECE
placement
result placement
result placement
result
Fig 1. Concept based segmentation

35
document. The set of sentences in each of these concept blocks are actually present is different
parts of the document.
Since DOM nodes are processed, the time taken for processing is less when compared to other
vector based and document graph[10] based models. The processing time required to build the
document graph is avoided in this approach.
4.2 Concept Based Segmentation Algorithm
The conceptual term frequency is an important factor in calculating the concept-based similarity
measure between topic blocks. The more frequent the concept appears in the verb argument
structures[4][9] of a sentence, the more conceptually similar the topic blocks are. Concept based
segmentation algorithm is described below:
Input : Web document di.
Output : Set of concept blocks{Cb1,..Cbn} of di,
Concept list of di, L={C1,..Cm}
Concept list of topic block tbi, Ctbi ={ck1,..ckm}, k=1..n
Step1: Mark all leaf nodes as individual micro blocks in the DOM tree.
Step2: Extend the border of the micro block to include all leaf nodes of the same tag to form a
topic block so as to have a set of topic blocks TB={tb1, tb2, …tbn}, TB⊂di.
Step3: Build concept list for all topic blocks TB ={tb1, tb2, …tbn}.
Topic block tbi is a set of sentences, tbi={si1, si2, ..sin1}, si⊂ tbi.
Sentence si is a string of concepts, si={ci1,ci2 . . . cim}, ci ⊂ si, if ci is a substring of si
Concept ci is a string of words, ci = {wi1,wi2, . . .wik },
where k : number of words in concept ci.
m: number of concepts generated from the verb argument structures in sentences
n1: total number of sentences in tbi
3.1: Ctbi, L are empty lists.
3.2: Build concept list of each sentence in tbi
3.2.1: si is a new sentence in tbi
3.2.2: Build concepts list Ci from si, Ci ={c1,c2,..cm}
3.2.3: Update concept list Ctbi of topic block tbi and L of document di
3.2.3.1: for each concept ci ⊂ Ci do
3.2.3.2: for each cj⊂ L, do
3.2.3.3: if (ci == cj) then
3.2.3.4: add freq(ci, si) to ctfi of ci
// freq(ci, si) returns the frequency of ci in the verb argument
// structures of si, added to conceptual term frequency of ci
3.2.3.5: else add new concept to Ctbi, L // added to both L and Ctbi
3.2.3.6: end if
3.2.3.7: end for
3.2.3.8: end for
3.3: Output the concepts list Ctbi
3.4: Output the concepts list L of document di.
Step4: The concept based similarity between topic blocks are measured by (1) .
2
2
2
1
1
1
2
1
2
1
2
1
1
)
,
( i
C
C
i
i
i
C
C
i
i ctfweight
ctf
ctfweight
ctf
tb
tb
Sim
tb
tb
tb
tb
×
−
×
−
= ∑
∑ ∩
⊂
∩
⊂
(1)
Where,

36
ctfi1, ctfi2 : Frequency of concept ci in tb1, tb2
i1,i2 : set of common conceptual terms between tb1, tb2
ctfweighti1,ctfweighti2 : Weight of concept ci with respect to topic blocks tb1, tb2
normalized by the frequency vectors of tb1, tb2,calculated as in (2)
( )
∑ =
=
m
k k
i
i
ctf
ctf
ctfweight
1
2
(2)
ctfweighti represents the importance of the concept ci with respect to the concept vector of tbi.
The concepts contributing more to the meaning of the sentences in the topic block occurs more
times in the verb argument structure of the sentence and in turn gets more weightage.
Step5: Merge the topic blocks having concept based similarity measure above the predefined
threshold α.
Concept block Cbk={set of topic blocks tbi}|
∀tbi, tbj∈Cbk, sim(tbi, tbj)> α, tbi⊂TB, tbj⊂TB, k=1..n
Step6: Output Concept blocks Cb1,Cb2,..Cbn
Similarity between topic blocks is measured by considering the commonly occurring concepts in
both topic blocks, tb1 and tb2. Frequency of these common terms and their topic block based
weightage are used to measure the similarity score and is normalized to the range 0 to 1.
The concept blocks of each URLs selected by the user are identified. Concept blocks of all URLs
in the repository can be identified during preprocessing stage itself. This will reduce the
computation complexity at run time.
The next section describes about generating the comparative summary on the fly at run time using
these concept blocks of the web document.
5. COMPARATIVE SUMMARY GENERATION
The architecture of the comparative summarization system is given in Fig.2. User enters the
generic query string (eg. Engineering College) through the search engine query interface. Search
engine identifies the relevant pages and present the search result in rank order. Then the specific
feature keywords based on which comparative summary is to be generated and the set of URLs
are obtained from the user.
The selected HTML files of the URLs are cleaned by removing unwanted HTML tags (like
META tag, ALIGN tag, etc.) which do not contribute much for further processing. Concept
blocks of these URLs which were already formed during preprocessing are utilized to generate
the summary.
The relevance of the concept blocks Cbi to the feature keywords f is measured by means of
similarity between the feature keyword string[4][9] and the concept list of each of the concept
blocks Cbi..
∑
∑
∈
∩
∈
=
Cbi
t
ti
Cbi
f
t
ti
i
f
Ctf
Ctf
Cb
f
sim
2
)
,
( (3)
Where,
f: set of feature keywords,f={f1,f2,..fn}

37
Cbi: Concept block i, for which the list of concept terms and their frequency were already
identified.
Simf (f, Cbi) : Similarity between feature keyword string f and Concept block Cbi
t : set of common terms between f and Concept list of Cbi
Ctfit : frequency of term t in Cbi
Simf(f,Cbi) is measured using the conceptual term frequency of the matching concepts of these
concept blocks and is normalized by the concept frequency vector of the concept block. The
concept block having maximum number of matching concept terms will get high score. The range
of Simf(f,Cbi) value lies between 0 and 1, and the similarity increases as this value increases.
Synonyms of the feature words and concept terms were taken into consideration for processing.
The concept block with maximum similarity is considered as the superset of the summary to be
generated.
The significance of each sentence in this concept block with respect to the query string is
measured. The sentences are considered in the descending order of their score. According to the
ratio of summarization required or the number of sentences required, the sentences are extracted
from these concept blocks to compose the HTML page for comparative summary.
5.1 Sentence Weight Calculation
Content of these concept blocks are ranked with respect to the query string and the feature
keywords, using (4). Sentence weight calculation considers[12] the number of occurrences of
query string, feature keywords and their distance and frequency, location of the sentence, tag in
which the text appears in the document, uppercase words,etc.
Fig. 2. Framework of comparative summarization system
Web
DB
Preprocessing
Use DOM tree & find
micro blocks
Clean the documents
Form topic blocks by
merging micro blocks
Find the concept lists of
topic blocks
Merge similar topic
blocks to form concept
blocks
At Run time
Compose & Display
Comparative summary
Display Search result
Search String
Search Engine
Enter feature keywords
Select set of URLs
Select top scoring sentences
from the concept blocks
based on keywords
Select relevant
concept blocks

38
)
(
)
(
)
( 1
)
1
(
i
n
i
l
tag
D
i
i
S
Len
W
W
e
k
W
S
W
∑
=
−
−
+
+
+
=
β
α
γ
(4)
Where,
W(ki ) :Weight of certain feature word or query word ki in sentence si
Len(si ) :number of words contained in the sentence.
Wtag :weight determined by the tag to which the words of sentence belong
BOLD, UNDERLINE, ITALICS, CAPTION, PARAGRAPH TITLE : 3
COLOR CHANGE :2
Wl :weight determined by the location of the Sentence with respect to the parent
node, set to 1 to the left most node and 0.5 to the right most node of a parent
D :average distance between feature words measured by counting the number
of other words in between the feature words
α , β and γ :adjusting parameters.
Table 1. Comparison between Search Engine Snippets, Mead and
Comparative Summarizer
Parameters
Search Engine
Snippets
MEAD
Summarizer
Comparative
summarizer
Method of summary
generation
Extraction Extraction Extraction
Techniques used
Occurrence of
query string
Lex rank,
Centroid
Position
DOM tree, Concept
based segmentation
Document type Web page Text Documents Web Pages
Single/multi
document
Single Multi Document Multi Document
User control No Control Less Control More Control
Satisfaction Index Low Medium Medium to high
Run time over head Less
Comparatively
high
Comparatively high
Usage
Gives a clue
about the
relevance of the
document
Short summary
is generated
Comparative
Summary is
generated which is
useful for decision
making
Retrieval Efficiency Need further Reduces time Reduces time and

39
browsing and
scanning
taken for
scanning entire
set of documents
to understand
the core concept
effort taken for
browsing and
scanning various
web pages to
extract the gist of it
The sentences having frequent occurrences of the feature keywords and enclosed in special tags
are given preference. Location based weight is assigned according to the location of the sentence
within its immediate parent node. The top scoring sentences are extracted and arranged based on
the hierarchical structure of the individual documents. The title or first sentence of the immediate
parent of the extracted sentence, is chosen as subtitles for a set of leaf node contents. (for
example, IT, CSE, ECE in our example). Hence the resulting summary will contain the
SECTION-wise summary sentences of a set of URLs chosen by the user for immediate
comparison. This is applicable to various decision making situations which require analysis of
various parameters from various sources.
6. COMPARISON WITH EXISTING SYSTEMS
Various feature of the proposed system and the search snippets and a bench mark text
summarization system are compare and is presented in Table 1. As there is no bench mark web
document summarizer is available the techniques used in MEAD[14] is compared with the
current system. The snippets are the set of sentences displayed by search engines as part of search
results along with URLs that are extracted from the web page. These are the line in which the
search string occurs in the web page.
As given in Table1, the snippets provide vague information which are not sufficient to guess the
usefulness of the target page and requires a complete scan of the content to capture the required
information.
MEAD provides a summary of set of text documents based on Lexrank and Centroid position
score. This prepares the summary of the core concept of the documents which might be modified
as per the query string.
This system makes use of concept based segmentation approach which give more importance to
conceptual terms contributing more to the meaning of the sentences. Hence this systems
performance is comparatively promising than the other summarization systems.
7. EXPERIMENTAL RESULTS AND DISCUSSION
The experimentation of this work was carried out with the real time dataset containing randomly
collected 200 web documents from internet related to the educational institutions, algorithms,
banking and household items.
Normally, the summarization systems are evaluated using intrinsic approach or extrinsic approach
[6][11]. Intrinsic approach directly analyzes the quality of the automatic summary through
comparing it with the abstract extracted by hand or generated from other different automatic
system. Extrinsic approach is a task-oriented approach, which measures the abstract quality
according to its contribution.
Intrinsic approach was utilized to conduct the experiment of this system. Users including one
engineering student, three naïve users and one expert level user were involved in the
experimentation process.

40
The evaluation feedback collected from the experimentation is listed in table 2. and is plotted
Fig. 3.
Table 2. Evaluation feedback measure
User# Query Feature Keywords Feed back
in 5 point
scale
#1 Engineering College,
Chennai
Placement, Recruiters 4
#2 Optimization
Algorithms
Efficiency, Time
Complexity
3.5
#3 Theme Park, Chennai Entry Fee, Games 4.2
#4 Banks Home loan rate, services 3.5
#5 Washing Machine Brands, cost, offers,
warranty
3.3
The query result user interface of the system is given in Fig. 4 through which the user selects the
URLs and enter feature words for comparative summary generation.
Fig. 3 User Feedback in 5 Point Scale

41
Fig. 4 Query result and URL selection interface
Summary extracted from these selected URLs are presented to the user as a comparative
summary given in Fig 5. This system produces the comparison information required for real time
decision making dynamically. The average user satisfaction index is 3.7.
Fig. 5 Comparative summary
Since DOM tree structure of the web documents are utilized to process the content and generate
the summary the time complexity involved is very less when compared to systems making use of
vector space model and document graph model.
8. CONCLUSION AND FUTURE WORK
This system focused on generating comparative summary from a set of URLs selected by the
user. Concept based segmentation is used to identify the relevant block of content in the
document and top scoring sentences are extracted, composed and displayed to the user. This
summary would definitely be helpful for the users to get the immediate summary and for decision
making.

42
The impact of usage of key words association, document graph model of documents on this
system and advanced text clustering techniques for summary generation can be done as a future
expansion to this system.
REFERENCES
[1] Chitra Pasupathi, Baskaran Ramachandran, Sarukesi, “Selection Based Comparative Summarization
Using Concept Based Segmentation”, CCIS 197, pp 655-664, Springer-Verlag Berlin Heidelberg,
2011.
[2] Md. Mohsin Ali , Monotosh Kumar Ghosh And Abdullah-Al-Mamun, “Multi-Document Text
Summarization: Simwithfirst Based Features And Sentence Co-Selection Based Evaluation”,
Proceedings Of IEEE International Conference On Future Computer And Communication, PP93-96,
2009.
[3] Ramakrishna Varadarajan, Vagelis Hristidis, And Tao Li, “Beyond Single-Page Web Search
Results”, IEEE Transactions On Knowledge And Data Engineering, Vol. 20, No. 3, PP411-424,
2008.
[4] Zhimin Chen, Jie Shen, “Research On Query-Based Automatic Summarization Of Webpage”, ISECS
International Colloquium On Computing, Communication, Control, And Management, pp173-176,
2009.
[5] Ahmed A. Mohamed, Sanguthevar Rajasekaran, “Improving Query-Based Summarization Using
Document Graphs”, Proceedings of IEEE International Symposium On Signal Processing And
Information Technology, PP408-410,2006
[6] Fu Lee Wa, Tak-Lam Wong, Aston Nai Hong Mak, “Organization Of Documents For Multiple
Document Summarization”, Proceedings Of The Seventh International Conference On Web-Based
Learning, IEEE, PP98-104,2008
[7] Elias Iosif and Alexandros Potamianos, “Unsupervised Semantic Similarity Computation between
Terms Using Web Documents”, IEEE TRANSACTIONS ON KNOWLEDGE AND DATA
ENGINEERING, VOL. 22, NO. 11, 2010, pp1637-1647.
[8] P. Kingsbury and M. Palmer, “Propbank: The Next Level of Treebank,” Proc. Workshop Treebanks
and Lexical Theories, 2003
[9] Shady Shehata, Fakhri Karray, Mohamed S. Kamel, “An Efficient Concept-Based Mining Model for
Enhancing Text Clustering”, IEEE Transactions On Knowledge And Data Engineering, VOL. 22,
NO. 10, pp1360-1371, 2010.
[10] Ohm Sornil, Kornnika Gree-Ut, “An Automatic Text Summarization Approach Using Content-Based
And Graph-Based Characteristics”, IEEE, 2006.
[11] Mani.I, “Summarization Evaluation: An Overview”, Proceedings Of The NTCIR Workshop Meeting
On Evaluation Of Chinese And Japanese Text Retrieval And Text Summarization, 2001
[12] Cem Aksoy, Ahmet Bugdayci, Tunay Gur, Ibrahim Uysal and Fazli Can, “Semantic Argument
Frequency-Based Multi-Document Summarization”, IEEE, 978-1-4244-5023-7/09, 2009, pp460-464.
[13] www.summarization.com/mead/

43
Authors
Ms. P. Chitra, Associate Professor, RMK Engineering College, Chennai, India received
B.E degree in Computer Science and Engineering from Bharathidasan University, Trichy,
India, in 1992, and the M.E degree in Computer Science and Engineering from Anna
University, Chennai, India, in 2006. Currently she is pursuing Ph.D programme in the
Department of Information and Communication Engineering at Anna University –
Chennai, India. She has presented 3 papers international conferences and authored 2
books. She has organized many Workshops and Seminars. She is a life membe r of ISTE
(Indian Society for Technical Education) and IACSIT (International Association of Computer Science and
Information Technology).
[[
Dr.R.Baskaran received his B.Tech in Electrical and Electronics Engineering, Masters
Degree in Computer Science and Engineering, Madras University and Doctor ate from
Anna University in the years 2000,2001, and 2007 respectively. He is now working as a
Assistant Professor in Department of Computer Science and Engineering, College of
Engineering Guindy, Anna University Chennai. His present research includes Networks,
Database and Image Processing. He has published 35+ papers in International, National
Journals and Conferences. He is a life member of Institution of Electronics and
Telecommunication Engineers (IETE), Indian Society for Technical Education (ISTE),
Computer Society of India (CSI) and International Association for Engineers.
Dr. K. Sarukesi, Vice Chancellor, Hindustan University has a very distinguished career
spanning over 40 years. He has a vast teaching experience in various universities in India
and abroad. He was awarded a common wealth scholarship by the association of common
wealth universities, London for doing Ph.D in UK. He completed his Ph.D from the
University of Warwick – U.K in the year 1982. His area of specialization is Technological
Information System. He worked as expert in various foreign universities. He has executed
number of consultancy projects. He has been honored and awarded commendations for his
work in the field of information technology by the government of Tamilnadu. He has published over 30
research papers in international conferences/journals and 40 National Conferences/journals.

Query Sensitive Comparative Summarization of Search Results Using Concept Based Segmentation

More Related Content

Similar to Query Sensitive Comparative Summarization of Search Results Using Concept Based Segmentation (20)

More from CSEIJJournal (20)

Recently uploaded (20)

Query Sensitive Comparative Summarization of Search Results Using Concept Based Segmentation