2014 Sixth International Conference on Computational Intelligence and Communication Networks

Opinion Mining and Summarization of Hotel

Reviews
Vijay B. Raut
Dept. of Information Technology D.D. Londhe
Pune Institute of Computer Technology Dept. of Information Technology
Pune, India Pune Institute of Computer Technology
vijayraut2009@gmail.com Pune, India
ddlondhe@pict.edu
Abstract— Everyday many users purchases product, book
travel tickets, buy goods and services through web. Users also reviews. There are many applications of opinion mining such
share their views about product, hotel, news, topic etc on web in as decision making, recommendation systems, feedback
the form of reviews, blogs, comments etc. Many users read analysis etc. It is one of the popular research areas in text
review information given on web to take decisions such as buying mining and natural language processing.
products, watching movie, going to restaurant etc. Reviews
contain user’s opinion about product, event or topic. It is difficult
for web users to read and understand contents from large
In this paper, we present opinion mining of hotel reviews
number of reviews. Important and useful information can be based on machine learning approach and SentiWordNet [10]
extracted from reviews through opinion mining and based approach. We also present sentence extraction based
summarization process. We presented machine learning and opinion summarization of hotel reviews. Section II contains
SentiWordNet based method for opinion mining from hotel related work, in section III we present our proposed system
reviews and sentence relevance score based method for opinion architecture, section IV contains experiments performed and
summarization of hotel reviews. We obtained about 87% of results obtained. Section V contains conclusion.
accuracy of hotel review classification as positive or negative
review by machine learning method. The classified and II. RELATED WORK
summarized hotel review information helps web users to
understand review contents easily in a short time. We presented a survey paper on opinion mining and
summarization [18]. Many researchers performed sentiment
Index Terms—Text Mining, Opinion Mining, Opinion analysis on different domains data such as movie [4], product
Summarization [13], and social network [3] [15]. Many researchers also
worked on summarization of reviews [1] [6] [7]. In a survey of
I. INTRODUCTION Pang and Lee [9] they presented concept of opinion mining, its
Due to growth of technology, many users access Internet as application and challenges involved it. Mikalai Tsytsarau,
a source of information. Many users do online transactions for Themis Palpanas [2] also presented a survey on opinion
shopping, booking tickets of travel, movies, hotels, etc. Users mining, in that survey they mentioned about different datasets,
share their experience about movie, product or topic on internet resources and work performed in this domain.
in the form of reviews [9]. Many websites such as
www.tripadvisor.com allow users to share their views about Sentiment analysis of reviews is performed by machine
hotel, restaurants, and tourist places [2]. Due to large number learning methods by many researchers [5] [8] [11]. Machine
of users express their thoughts about hotels in reviews, there is learning method uses different classification algorithms for
need of mining user generated contents present in reviews. opinion classification. Classification algorithms such as Naive
Web users also read and use review information for decision Bayes [11], SVM [4], and Decision Tree [1] work well for
making about product, movie, and hotel etc. It is difficult for a review text. Other methods such as resource based opinion
user to read and understand all reviews about a particular hotel. mining [6], [7] also proposed by different researchers. It
Relevant and important information about hotel should be depends upon resources such as SentiWordNet [10], WordNet
fetched from reviews and presented to user in summarized Affect [14] which contain polarity values of word for sentiment
manner. classification.

Opinion mining is a process of identifying user’s opinion Some researchers suggested Text summarization
about movie, hotel, and product from reviews [2]. Opinion approaches [6] for opinion summarization of review text.
mining is classification of user‘s expressed opinion into Authors also proposed aspect based summarization of product
positive or negative polarity. Opinion summarization is a reviews in [11] [12]. In [1] author presented aspects and
process of representation of review information in short and related word phrases as summary. Many researchers used
summarized form [6]. It also involves selecting important Latent semantic analysis (LSA) for opinion summarization of
aspects and representing related expressed opinions from user reviews [4] [7].

978-1-4799-6929-6/14 $31.00 © 2014 IEEE 557

556
DOI 10.1109/CICN.2014.126
10.1109/.126
 Hotel review websites Pre-processing
(www.tripadvisor.com)
Sentence Segmentation

Review Retrieval Tokenization

Opinion Mining
(Web crawling)
Machine Learning POS Tagging
Classifiers / SentiWordNet
based method
Review text

Opinion Summarization

Sentence scoring
(Relevance score)
Positive reviews Negative reviews

Summary generation

Summary

Figure 1: Proposed System Architecture.

III. OVERVIEW AND ARCHITECTURE OF SYSTEM Naive Bayes, Support Vector Machine and Decision Tree.
The prepared classifier models of different classifiers are
Opinion mining is identification of user’s opinion about evaluated for their accuracy.
particular topic from reviews. It is classification of a review
text as positive or negative opinion polarity review. Opinion 2) SentiWordNet based method
summarization is a process of finding most important aspects
about topic and related opinion sentences from reviews to SentiWordNet [10] is one of the popular lexical resources
represent a summary. Our proposed architecture performs for sentiment analysis. SentiWordNet is a database that
opinion mining and summarization of hotel reviews is shown contains word with its polarity score for positive or negative
in Figure 1. sentiment based on its part of speech. By calculating scores of
different words in review we get polarity of review text. We
The proposed system consist of three modules review text also need to handle negation words such as ‘not’, ‘don’t’,
retrieval, classification and summarization. Reviews about ’never’ etc that affect the polarity of sentence. If S is polarity
hotel are retrieved from review websites such as of document, for a word w sentiment score is calculated by
www.tripadvisor.com by web crawling techniques. Review text function sentiment (w) and a function modifier (w) which
is classified as positive or negative review using machine handles word position or negation in sentence [2]. General
learning classifiers or SentiWordNet based algorithm. equation to identify sentiment polarity of review text
Classified review text is pre-processed and sentence scores are document D is [2]
calculated. Finally most informative and context relevant
sentences represented in summary. σ௪‫א‬஽ ‫ݐ݊݁݉݅ݐ݊݁ݏ‬ሺ‫ݓ‬ሻ ȉ ݉‫ݎ݂݁݅݅݀݋‬ሺ‫ݓ‬ሻ
ܵሺ‫ܦ‬ሻ ൌ ሺʹሻ
A. Opinion mining of hotel reviews σ ‫ݐ݊݁݉݅ݐ݊݁ݏ‬ሺ‫ݓ‬ሻ
Opinion mining is a text classification problem where B. Opinion summarization of hotel reviews
review text document is classified into classes as positive or
negative opinion review. Machine learning approach [4] [11] For opinion summarization of reviews we are using
and resource based approach [6] [7] for opinion classification sentence extraction method. Sentence extraction is an approach
of review text can be used. for text summarization. In this method sentences are filtered
and most informative sentences are selected from document to
1) Machine learning approach represent a summary. To find summary from large number of
reviews we are selecting those sentence which are informative
Machine learning approach is supervised text and most relevant to the context. Review sentences are scored
classification, where algorithms are trained on sample labeled and sorted based on relevance score [6]. Relevance score is
review text to build a classifier model. A trained model of calculated by following formula [6]. It is proposed in [6] by
classifier is then used for categorization of new test reviews. Lloret et al.
We trained and tested machine learning classifiers such as

557
558
 TABLE 1: CLASSIFICATION RESULTS V. CONCLUSION
Number of 1000 (500-positive, 500-negative)
reviews This paper presented opinion mining and summarization of
Training hotel reviews on the web. For opinion classification of hotel
Number of 1000 (500-positive, 500-negative) reviews we used machine learning algorithms and
reviews
Testing
SentiWordNet based method. We obtained about 87%
Classifier NB SVM Decision SentiWordNet
accuracy for classification of hotel review text as positive or
Tree method negative opinion polarity review by machine learning Naïve
Correctly 880 835 784 876 Bayes algorithm. For opinion summarization, we used term
classified frequency and relevance scoring method to represent most
Accuracy 88% 83.5% 78.4% 87.6% informative sentences in summary. This classified and
Precision 0.895 0.844 0.786 0.90
Recall 0.890 0.835 0.784 0.876
summarized review information will assist users in decision
F-measure 0.879 0.834 0.784 0.888 making about hotels.

ͳ REFERENCES
ܴܵ௜ ൌ ෍ ‫݂ݐ‬௪ ሺ͵ሻ
ܰܲ௜ [1] Jingjing Liu, Stephanie Seneff, and Victor Zue,
௪‫א‬ே௉ "Harvesting and Summarizing User-Generated Content for
Advanced Speech-Based HCI", IEEE Journal of Selected
Where NPi is number of noun phrases contained in sentence i Topics in Signal Processing, Vol. 6, No. 8, Dec 2012,
and‫݂ݐ‬௪ is frequency of word w belongs to that noun phrase pp.982-992.
[6]. [2] Mikalai Tsytsarau, Themis Palpanas "Survey on mining
subjective data on the web", Data Mining Knowledge
IV. EXPERIMENTS AND RESULTS Discovery, Springer 2012, pp.478-514.
We evaluated our approach of opinion classification based [3] Alvaro Ortigosa, José M. Martín, Rosa M. Carro,
on parameters such as Precision, Recall and F-measure. For "Sentiment analysis in Facebook and its application to e-
opinion classification of hotel reviews we used tripadvisor learning", Computers in Human Behavior Journal Elsevier
dataset [17]. We selected and labelled 1000 reviews from it for 2013.
training and another 1000 reviews for testing. [4] Chien-Liang Liu, Wen-Hoar Hsaio, Chia-Hoang Lee, Gen-
Chi Lu, and Emery Jou “Movie Rating and Review
Summarization in Mobile Environment”, IEEE
We tested different machine learning classification Transactions on Systems, Man, and Cybernetics-Part C:
algorithms for opinion classification of review using weka Applications and Reviews, Vol. 42, No. 3, May 2012,
library [16]. We also tested SentiWordNet based approach on pp.397-406.
test reviews. Table 1 show results obtained through our [5] Aditya Joshi, Balamurali A. R., Pushpak Bhattacharyya "A
experiments of hotel review text classification. Graph of Fall-back Strategy for Sentiment Analysis in Hindi a Case
accuracy of different classifiers is shown in Figure 2. Study" Proceedings of ICON 2010: 8th International
Conference on Natural Language Processing, Macmillan
Publishers, India.
Accuracy of Classifiers [6] Elena Lloret, Alexandra Balahur, José M. Gómez, Andrés
100% Montoyo, Manuel Palomar, "Towards a unified framework
90% for opinion retrieval, mining and summarization" Journal of
80% Intelligent Information Systems Springer 2012, pp.711-
70% 747.
60%
50% [7] Alexandra Balahur, Mijail Kabadjov, Josef Steinberger,
40% Ralf Steinberger, Andrés Montoyo, "Challenges and
30%
20% solutions in the opinion summarization", Journal of
10% Intelligent Information Systems Springer 2012, pp.375-
0% 398.
[8] Alexandra Trilla, Francesc Alias "Sentence-Based
Sentiment Analysis for Expressive Text-to-Speech", IEEE
Transactions on Audio, Speech, and Language Processing,
Vol. 21, No. 2, February 2013, pp.223-233.
[9] Bo Pang, Lillian Lee, "Opinion Mining and Sentiment
Analysis", Foundations and Trends in Information
Retrieval Vol. 2, Nos. 1–2 (2008).
Figure 2: Accuracy of Different Classifiers [10] Esuli, A., & Sebastiani, F. (2006). “SentiWordNet: A
publicly available resource for opinion mining”. In
Proceedings of the 6th international conference on
Language Resources and Evaluation (LREC’06), pp.417–
422.

558
559
[11] Pang B, Lee L, Vaithyanathan S. “Thumbs up? Sentiment [14] Fellbaum, C. (1998). WordNet: An electronic lexical
classification using machine learning techniques”. database. The MIT Press.
Proceedings of the Conference on Empirical Methods in [15] Aditya Joshi, Balamurali A. R, Pushpak Bhattacharyya,
Natural Language Processing (EMNLP) 2002. Rajat Mohanty, "C-Feel-It: A Sentiment Analyzer for
[12] Dave K, Lawrence S, Pennock D. “Mining the peanut Micro-blogs", Proceedings of the ACL-HLT 2011, pp.127-
gallery: opinion extraction and semantic classification of 132.
product reviews”. Proceedings of the 12th international [16] Weka 3: Data Mining Software in Java, available at
conference on World Wide Web, ACM, New York, NY, http://www.cs.waikato.ac.nz/ml/weka
USA, WWW’03.
[17] Tripadvisor Review Dataset available at
[13] Hu M, Liu B. "Mining and summarizing customer http://sifaka.cs.uiuc.edu/~wang296/Data/index.html
reviews". Proceedings of the 10th ACM SIGKDD
[18] Vijay B. Raut, D.D. Londhe "Survey on Opinion Mining
international conference on knowledge discovery and data
and Summarization of User Reviews on Web",
mining. ACM, New York, NY, USA, KDD 2004.
International Journal of Computer Science and Information
Technologies, Vol. 5 (2) , 2014, pp. 1026-1030

559
560