A FRAMEWORK FOR SUMMARIZATION OF ONLINE OPINION USING WEIGHTING SCHEME

Advanced Computational Intelligence: An International Journal (ACII), Vol.2, No.3, July 2015
DOI:10.5121/acii.2015.2302 13
A FRAMEWORK FOR SUMMARIZATION OF ONLINE
OPINION USING WEIGHTING SCHEME
Mugdha More1
and Bharat Tidke2
1
Department of Computer Engineering, Savitribai Phule Pune University, Khopi, Pune-
412 205, India
2
Department of Computer Engineering, Savitribai Phule Pune University, Khopi, Pune-
412 205, India
ABSTRACT
Recently there is wide use of social media includes various opinion sites, complaints sites, government
sites, question-answering sites, etc. through which customer get services, opinion, information, etc. but
because of this there is more and more use of these social media right now so huge amount of data will be
created, from this huge data people get confused while taking any decision about particular problem or
services. For example, customer wants to purchase a product at that time he/she want the previous
customer feedback or opinion about that product. But if there is lots of opinion available for particular
product then that customer get confused while taking decision whether purchase that product or not. In this
case there is a need of summarization concept means that only show the short and concise manner
summary about service or product so that customer or organization easily understand and able to take
right decision fast. Our proposed framework creating such summary which contain three main phases or
steps. Firstly preprocessing is done in that stop words are removed and stemming is performed. In second
phase identify frequent features using two techniques weight constraint and association rule and at the last
phase it find semantics and generate the summary so that customer will able to take step without confusion.
KEYWORDS
Association, Opinion, Preprocessing, Weighting, Summary
1.INTRODUCTION
People expose their thoughts, things with different sites and also they can communicate with each
other through social media [1] [2]. They also give their views about particular services or
products through web in their own language but each human thinking is different so various types
of opinions are available for single thing [1] [2] [3]. From this opinion customer or organization
get various way of thinking or dimensions of particular product or services but all dimensions
may not be useful or may be some dimensions are fake hence automatically the opinion are fake
so there is a need of identification of such opinion called opinion spam [3] [4]. Also there are
some opinions which are duplicate or near duplicate [2] [5] [6]. Existing features focus on
classification, summarization, spam detection, association, sentiment analysis, semantic analysis,
opinion mining, etc. for identifying true opinions from fake, duplicate and near duplicate reviews
or opinion [4]. Classification means categorization of whole data into different subgroups
considering some constraints or rules. For example if there is a group of people and we have to
classify them then we take constraint as a gender hence we get two group of people female and
male [1] [2]. Summarization is nothing but summary of data i.e. short and important data from
whole data. If there is ten pages data available then from summarization we get two to three page

14
proper and important data which are useful. Summarization is closely related to opinion mining
concept [2]. Opinion mining is a technique which analyzes the data from different perspective and
summarizes it into knowledgeable information [3]. Opinion mining is also called as sentiment
analysis is nothing but analyze the data with its sentiments, also it is a study of emotions,
appraisals, opinion of peoples [7]. Association technique is a rules created for frequent features
identification which having some relationships between each other [2] [3] [8]. But while handling
such techniques various difficulties are occurred like word sense disambiguation, negation
handling, etc. because opinions are in natural language and all this techniques handle such natural
language opinions [7]. There is need of deep understanding and knowledge about language rules
like implicit or explicit, regular or irregular, syntactic or semantic, etc. for handling such
problems [9]. Various companies, organization, markets, businesses required such techniques for
their company or business improvements, decision about services, etc. In this paper we propose a
framework containing three phases for creating concise summary which minimizes the customer
or organization confusion.
2.RELATED WORK
Our work is related to sentiment analysis, opinion mining and opinion summarization. Existing
research is on various domains and on various techniques. G. Vinodhini and RM. Chandrasekaran
have been focused on performance of classification model and different features that shows
sentiments [10]. Bing Liu and Lei Zhang proposed unified framework and identify mining tasks
and also find out techniques and issues of opinions [11]. Polarity classification accuracy is
increased through minimum cuts framework which is proposed by Bo Pong and Lillian Lee [12].
Bing Liu represents the problem of opinion mining in sentence, document and aspect-level [13].
Also there is a technique which finds opinions with its holder and topic introduced by Su Nam
Kim et al. [14]. Also there is a technique for improving performance on opinion mining using
syntactic dependency relationship and technique for converting dependency relation triples to
composite features which are described by Mahesh Joshi et al. [15]. Eric Breck et al. proposed
opinion expression identification approach with the help of conditional random field and also
evaluate that approach [16]. Some research is on identification of opinion boundary and its
intensity without exploiting hierarchical structure by Yejin Choi et al. [17]. For opinion statement
analysis rule based approach is proposed by Alena Neviarouskaya et al. They do classification of
verb and create rule for each verb class [18]. Summarization of huge data is done by using graph-
based summarization framework called Opinosis by Kavita Ganeson [19]. The method for feature
mining and creating summary is introduced by Minqing Hu and Bing Liu. They also find out
difficulties while creating such summary [20]. By using summarization method and ranking
technique Yihong Gong and Xin Liu represent the text summarization through which they
increase the performance [21]. The overview of research done in sentiment analysis and
summarization is shown in Table 1 with data set.
Table 1. Existing research technique with its features and dataset.
Author Technique Feature Dataset
G. Vinodhini et
al.
Sentiment Analysis Sentiment method,
technique, challenges,
classification
Movie Review,
Product Review
Bing Liu et al. Opinion Mining,
Sentiment Analysis
Framework to unify
research directions
Product Review
Bo Pong et al. Sentiment polarity,
Minimum Cuts in
Graphs
Machine Learning
Method, Text
Categorization Techniques
Movie Review
Su Nam Kim et Semantic role labelling Target word, Phrase Type, FrameNet, The

15
al. method Head word, Parse tree
path, Position, Voice,
Frame name
New York Times,
BBC News, etc.
Mahesh Joshi
et al.
Syntactic dependency
relation triples
Composite Back-off, Full
Back-off, Ngram Back-off
Amazon.com,
CNET.com
Eric Breck et
al.
CRF Lexical, Syntactic,
Dictionary based
MPQA Corpus
Yejin Choi et
al.
Hierarchical parameter
sharing technique, CRF
Per-token, Transition MPQA Corpus
Alena
Neviarouskaya
et al.
Rule-based approach,
Verb classification
Opinion Type, Strength,
Confidence level, Verb
class
VerbNet
Kavita Ganesan
et al.
Abstractive Summary Opinosis Framework Hotel Review,
Car Review,
Product Review
Minqing Hu et
al.
Semantic orientation,
Term Phrases
Bootstrapping Technique,
Feature Extraction Method
Amazon Reviews
Yihong Gong
et al.
Semantically Important
Sentences, Improve
Summarization
Performance
Standard IR Model, Latent
Semantic Analysis, VSM
Weighting Schemes
CNN Worldview
News Program
3.PROPOSED FRAMEWORK
Figure 1 gives proposed framework for online opinion summarization. The inputs to the
framework are date and time, product name and review of that product. The output is summary of
the review in short and concise manner. The system perform the summarization in three steps: (1)
Product feature get mined which is given by customer; (2) Identify frequent features in each
opinion sentence in each review; (3) Finding out whether feature are opinionated and also identify
orientation of the words and finally summary will be created. Multiple sub-steps are involved for
performing these three steps.
3.1.Dataset
We conducted our experiments using customer reviews of hotels, tourism, hard disk and baby
product. Hotels include reviews of two hotels, Hard disk contain the reviews of two companies
product, tourism have two location and in baby product two companies product are included. The
reviews were collected from Amazon.com. Product in these sites has large number of reviews.
3.2.Pre-processing
The framework first downloads all reviews and put them in the review database. We use the
parser to parse each review to split text into sentence to produce part of speech tags for each
word, reason behind is that the original opinion is in unstructured format means that they are
written in users natural language [22]. Preprocessing task is also performed on these documents.
It removes stop word or unimportant words in filtration step for that it maintain a stop word list.
Words are matched with this list, if match found then remove that word from opinion and if new
word is found as a stop word then add that word in the stop word list. It mainly includes a, an, the,
etc. words that was only used for grammatically complete the sentence. After that stemming is

16
performed, in this process root form of inflected words is find out and also eliminate the prefixes
and suffixes of the words. If there is a words like developed, written then after stemming the
words like develop and write.
Figure 1. Feature Based Online Opinion Summarization Framework
3.3.Frequent Feature Identification
The In this step we identify product feature which is present in opinion that express by users.
Before going to actual technique we firstly introduced that how difficult to extract feature from
opinion which are in natural language. Some opinions are directly express on entity explicitly so
that extract feature from these is not so much difficult. For example, “Sound quality of Samsung
phone is good” but there are some opinion which expressed implicitly so it difficult to find out
features. For example, “Samsung phone does not fit in pocket” means here user indirectly talks
about size of the phone. In our work, we used two techniques for identifying frequent features.
Firstly we use weighting scheme on filtered and stemmed XML document to index the feature.
These process is possible by manually and automatically also but here we use TF-IDF weighting
scheme for assigning weights to keywords [8]. The formula for TF-IDF weighting scheme is as
follows:

17
w, = , =
, ∗ log
||

, ≥ 1
0 , = 0

where, w(a, b) ≥ 0, , number of terms tb occurs in document da, denotes the number of
document in collection C in which tb occurs at least once and |C| denotes the number of document
in collection C.
After indexing to features by weighting schemes, association rule is applied on it to find all
frequent item sets. Association mining is nothing but identifying relationship between the words
hence we get frequent feature set [6]. In our case, we apply association on frequent features to get
more accurate frequent feature set. These frequent features are getting by calculating support
formula. This support must be greater than minimum support which is given by user. Support
formula is as follows:
Support (wa, wb) =
!#$ %#! #' (),(
*# + !, -$ #' .#%!,-
where, wa, wb are the frequent features. We combine these two techniques for getting more
accuracy in identifying frequent features.
3.4.Finding Semantics
In this step, firstly remember one thing is that we are only interested in customer opinion. In the
first sub-step we extract the opinion word that express the subjective opinion. In some cases
presence of adjectives is more useful for identifying subjective opinion. In previous step we
assign the tags to word using part-of-speech tag technique. With the help of these tags and its
position in the sentences we get adjective, adverb, noun and all so extracting the opinion words
from these words are easy. Identification of a semantic orientation of a word is useful for
identifying semantic orientation of sentences. Each word having semantic orientation which is
directed to its semantic group. If the word shows positive orientation that means word gives
desirable state and if the word shows negative orientation that means word gives undesirable
state. To predict the semantic orientation we maintain the synonym set and antonym set. Each
new word shows or has either positive or negative orientation but for predicting that orientation
we have to search its synonym or antonym set so according to that we get its orientation. Initially
we manually create a list called seed list by considering very common adjectives like good, nice,
fast, etc. as a positive word and bad, slow as a negative word. For predicting the orientation of
that word we search that word in WordNet. If found the orientation then add it in to our list.
According to this all procedure semantic orientation of a word is find out.
3.5.Summary
After performing all previous steps, we are ready to generate feature based online opinion
summary which is short and in concise manner. Previously we finding out frequent features after
that we got its orientation for particular product. After all this we get frequent opinion about
particular product then we convert that all frequent opinion into single well defined opinion
without changing its meaning so that customer easily understand and easily get the meaning of
opinion in short time.

18
4.EXPERIMENTAL EVALUATION
A Framework called feature based online opinion summarization based on proposed techniques
has been implemented in Java. We conducted our experiment using the customer review of four
products having different company. For each product, firstly we crawled and download the
reviews. After that, by using parser we generate part of speech tags then to perform
summarization our framework is applied on it. User interface of our framework is shown in figure
2.
Figure 1. GUI for summarization framework.
In that we provide a facility for selecting a file which contains all reviews. Also we can select the
category whatever we want. We show these all functionality in following figure 3.
Figure 2. Functionality in the framework.
After clean and sort we see the results in session tab called user wise session which is shown in
following figure 4.

19
Figure 3. User wise session of a dataset.
To apply the apriori technique, we provide apriori functionality in that we manually give the
support and calculate it. The result is in apriori tab with the time required for calculation i.e.
mining time in milliseconds with the efficiency chart and theoretical representation of this chart is
in likes tab which is shown in following figure 5 and figure 6 respectively.
Figure 4. Apriori result with efficiency chart.

20
Figure 5. Theoretical representation of efficiency chart.
5.CONCLUSIONS
In this paper, we study the various methods and techniques related to different domains like
opinion summarization, sentiment analysis, etc. This framework collects the different set of
opinions and applies all techniques which we already explained for getting summary. This
summary is shown in the form of graphs and also in text which are easily understandable.
In future work, we mainly focus on words which gives different orientation in different situations.
Also we are trying to increase accuracy in finding out features which is implicit. We also plan to
visualize our final result in more informative and accurate text summary.
REFERENCES
[1] B. A. Tidke, R. G. Mehta, D. P. Rana, “A Novel Approach for High Dimensional Data Clustering”, in
International Journal of Engineering Science and Advanced Technology (IJESAT) ISSN 22503676
Vol.02(3) May-Jun 2012.
[2] Mugdha More, Bharat Tidke, Social media online opinion summarization using ensemble technique,
Pervasive Computing (ICPC), 2015 International Conference on , vol., no., pp.1,6, 8-10 Jan. 2015.
[3] Nitin Jindal and Bing Liu, “Review Spam Detection”, Proceedings of 16th International World Wide
Web conference, WWW '07. Banff, Alberta, Canada 2007.
[4] Nitin Jindal and Bing Liu, Opinion Spam and Analysis, Proceedings of First ACM International
Conference on Web Search and Data Mining (WSDM-2008), Feb 11-12, 2008, Stanford University,
Stanford, California, USA.
[5] Snehal Dixit and A. J. Agrawal, “Survey On Review Spam Detection”, International Journal of
Computer Communication Technology ISSN, 2013.
[6] Minqing Hu and Bing Liu, Opinion Extraction and Summarization on the Web, Proceedings of 21st
National Conference on Artificial Intelligence (AAAI-2006, Nectar paper), July 16.20, 2006, Boston,
Massachusetts, USA.

21
[7] Qingliang Miao, Qiudan Li, Ruwei Dai, “AMAZING: A sentiment mining and retrieval system”,
Expert Systems with Applications 36 (2009) 7192–7198.
[8] Hany Mahgoub, Dietmar Rosner, Nabil Ismail and Fawzy Torkey, “A Text Mining Technique Using
Association Rules Extraction”, World Academy of Science, Engineering and Technology, Vol: 2 2008-
06-21.
[9] Long-Sheng Chen, Cheng-Hsiang Liu, Hui-Ju Chiu, “A neural network based approach for sentiment
classification in the blogosphere”, Journal of Informetrics 5 (2011) 313–322.
[10] G Vinodhini, RM Chandrasekaran, “Sentiment analysis and opinion mining: a survey”, International
Journal of Advanced Research in Computer Science and Software Engineering, Volume 2, Issue 6,
June 2012.
[11] Bing Liu and Lei Zhang, “A Survey of Opinion Mining and Sentiment Analysis”.
[12] Bo Pang and Lillian Lee, “A sentimental education: Sentiment analysis using subjectivity
summarization based on minimum cuts”, 2004 ACL.
[13] Bing Liu, “Sentiment Analysis and Opinion Mining”, Morgan Claypool Publishers, 2012.
[14] Mahesh Joshi and Carolyn Penstein-Rose, “Generalizing Dependency Features for Opinion Mining”,
In ACL, 313–316, 2009.
[15] Xiangyang Lan, Stefan Roth, Daniel Huttenlocher and Michael J. Black, “Efficient Belief Propagation
with Learned Higher-Order Markov Random Fields”, In ECCV, 269–282, 2006.
[16] Eric Breck, Yejin Choi and Claire Cardie, “Identifying Expressions of Opinion in Context”, In IJCAI,
2683–2688, 2007.
[17] Yejin Choi and Claire Cardie, “Hierarchical Sequential Learning for Extracting Opinions and Their
Attributes”, In ACL, 269–274, 2010.
[18] Alena Neviarouskaya, Helmut Prendinger and Mitsuru Ishizuka, “Semantically distinct verb classes
involved in sentiment analysis”, In IADIS, 27–35, 2009.
[19] Ganesan Kavita, Zhai ChengXiang and Han Jiawei , “Opinosis: A Graph Based Approach to
Abstractive Summarization of Highly Redundant Opinions”, Proceedings of the 23rd International
Conference on Computational Linguistics (COLING '10), (2010).
[20] Minqing Hu and Bing Liu, Mining and summarizing customer reviews, Proceedings of the ACM
SIGKDD International Conference on Knowledge Discovery Data Mining (KDD-2004, full paper),
Seattle, Washington, USA, Aug 22-25, 2004.
[21] Yihong Gong and Xin Liu, “Generic Text Summarization using Relevance Measure and Latent
Semantic Analysis”, SIGIR‟01, September 9-12-2001, New Orleans, Louisiana, USA.
[22] Chris Manning and Hinrich Schütze, “Foundations of Statistical Natural Language Processing”, MIT
Press. Cambridge, MA: May 1999.

A FRAMEWORK FOR SUMMARIZATION OF ONLINE OPINION USING WEIGHTING SCHEME

More Related Content

Similar to A FRAMEWORK FOR SUMMARIZATION OF ONLINE OPINION USING WEIGHTING SCHEME (20)

Recently uploaded (20)

A FRAMEWORK FOR SUMMARIZATION OF ONLINE OPINION USING WEIGHTING SCHEME