2010 Fifth International Conference on Internet and Web Applications and Services

BlogMiner: Web Blog Mining Application for Classification of Movie Reviews

Arzu Baloglu Mehmet S. Aktas

Marmara University, Engineering Faculty TUBITAK, The National Research Institute of
Computer Engineering Department Electronics and Cryptology (UEKAE), Information
Istanbul, Turkey Technologies Institute, Kocaeli, Turkey
E-mail: [email protected] E-mail: [email protected]

Abstract—With the increasing use of Web 2.0 platforms The organization of this paper is as follows. Section 2
such as Web Blogs, discussion forums, Wikis, and discusses the literature survey. Sections 3-4 outline the
various other types of social media, people began to proposed approach and the system architecture. Section 5
share their experiences and opinions about products or addresses the evaluation study. Section 6 concludes the paper
services on the World Wide Web. Web Blogs have thus with a summary and analysis of results.
become an important source of information. In turn,
great interest in blog mining has arisen, specifically due II. LITERATURE SURVEY
to its potential applications, such as in opinion or review In recent years, there has been a huge burst of research
search engine applications the ability to collect and activity in the areas of sentiment analysis and opinion
analyze data. In this study, we introduce an architecture, mining. Earlier studies focused mostly on interpretation of
implementation, and evaluation of a Web blog mining narrative, point of view in text [5-10]. The widespread
application, called the BlogMiner, which extracts and awareness of the research problems in sentiment analysis and
classifies people’s opinions and emotions (or sentiment) opinion mining has increased with the rise of machine
from the contents of weblogs about movie reviews. learning methods in natural language processing and
information retrieval; of the availability of datasets for
Keywords: blog mining, opinion mining, blog crawler, machine learning algorithms to be trained on (due to the
web blog mining blossoming of the World Wide Web); and, specifically, of
the development of review-aggregation Web sites.
I. INTRODUCTION Reference [3] describes a sentiment classification
The world’s biggest library, the World Wide Web, is application that uses phrase patterns to classify opinions. In
getting fed with data by every internet user around the world. this study, at the document classification phase, the authors
People share their ideas, interests, emotions, experiences, add tags to certain words in the text, and then match the tags
and knowledge with others via the internet every day. Thus, within a sentence with predefined phrase patterns to get the
mining opinions of people on the Web is an important area sentiment orientation of the sentence under consideration.
of research investigation [1]. Next, they take into account the sentiment orientation of each
Sociologists have used many different ways to recognize sentence and classify the text according to the most repeated
people’s natural interests, their aims, and their preferences. sentiment.
In order to grab people’s ideas from their sharing over the Reference [4] describes a sentiment miner that extracts
Web, the most efficient way is to mine people’s diaries or sentiment (or opinions) that people express about a subject,
books (in other words their blogs). This study introduces a such as a company, brand, or product name. In this study, the
system that is designed to mine ideas to understand the point authors design the sentiment miner with the following
of views of a web community. challenge in mind: Not only is the overall opinion about a
With increasing usage of the internet, blogging and blog topic, but also the sentiment about individual aspects of the
pages are growing rapidly. Blog pages have become the most topic essential information of interest. The reason for this is
popular means to express one’s personal opinions. By the that the document level sentiment classification fails to
end of 2008, there were 133 million blogs on the global detect sentiment about individual aspects of the topic. Thus
Internet, which are indexed by Technorati [2]. in the author’s study, the sentiment miner analyzes
Mining opinions from Web pages involves several grammatical sentence structures and phrases based on natural
challenges. For example, these opinions, or review data, have language processing (NLP) techniques and detects, for each
to be crawled from Web sites and then separated from non- occurrence of a known topic spot, the sentiment specifically
review data [8]. about the topic. With these characteristics the proposed NLP
This study proposes a system that extracts review data, of based sentiment mining system, described in [4], achieved
movies from blogs. It introduces an architecture and high quality results (∼90% of accuracy) on various datasets
implementation of the system in detail. It also explains a including online review articles and the general Web pages
classification of the review data. and news articles. The feature extraction algorithm, proposed
by [4], successfully identified topic related feature terms

978-0-7695-4022-1/10 $26.00 © 2010 IEEE 77

DOI 10.1109/ICIW.2010.19

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
from online review articles, enabling sentiment analysis at feature words and opinion words in a sentence; determining
finer granularity. the class of feature word and the polarity of opinion word;
Reference [5] describes an application on sentiment identifying the relevant opinion word(s) and then obtaining
classification with review extraction. This approach extracts some valid feature-opinion pairs; producing a summary
the review expressions on specific subjects and attaches a using the discovered information. The authors use WordNet
sentiment tag and weight to each expression. Then, it to generate a keyword list for finding features and opinions.
calculates the sentiment indicator of each tag by Grammatical rules between feature words and opinion words
accumulating the weights of all the expressions are then applied to identify the valid feature-opinion pairs.
corresponding to a tag. Next, it uses a classifier to predict the Finally, the authors re-organize the sentences according to
sentiment label of the text. In this study, the authors use on- the extracted feature-opinion pairs to generate the summary.
line documents to test the performance of the proposed The objective of this study is to generate automatically a
application. The experimental documents cover two feature class-based summary for arbitrary online movie
domains: politics and religion. The experiments within those reviews. Experimental results show that this method has an
domains achieve accuracy between %85 and %95. average precision of %65 approximately. In addition, with
Reference [6] describes a method of opinion mining to this approach, it is easy to generate a summary with movie-
help e-learning systems know the users’ opinions on the related people names as the sub-headlines.
course-wares and teachers of the e-learning system and to In this study, we propose a work, which is most similar
help improve the services. In this study, the authors develop with the work described in [9]. Our approach differs from [9]
an opinion mining system for e-learning reviews. The goal of in the way we calculate sentiment orientation of the movie
this system is to extract and summarize the opinions and reviews from the blogs. The previous work focused on a
reviews, and determine whether these reviews and opinions constant dataset, while the proposed approach crawl the
are positive or negative. This study divides the whole task dataset from the blogs. In turn, this is used to calculate movie
into four subtasks: expression identification, opinion scores. We discuss our approach in the next section in detail.
determination, content-value pair identification, and
sentiment analysis. The authors achieve following precisions III. APPROACH
for these subtasks respectively: %94, %84.2, %80.9 and
%92.6. A. Overview
Reference [7] describes a sentiment mining and retrieval In this section, we briefly describe the techniques and
system called Amazing. The authors introduce a ranking goals of this study and what we aim to succeed as a result.
mechanism, which is different from a general web search This study is categorized under three phases. The first phase
engine since it utilizes the quality of each review rather than is the crawling phase, in which data is gathered from Web
the link structures for generating review authorities. In this blogs. The second phase is the analyzing phase, in which the
system, the most important aspect is that the authors data is parsed, processed and analyzed to extract useful
incorporate the temporal dimension information into the information. The third phase is the visualization phase, in
ranking mechanism, and make use of temporal opinion which the information is visualized to better understand the
quality and relevance in ranking review sentences. This results. More details of the system architecture are explained
study monitors the changing trends of customer reviews in in the system architecture section (IV).
time and visualizes the changing trends of positive and B. Problem Definition
negative opinion respectively. It then generates a visual
comparison between positive and negative evaluation of a Web blogs are full of un-indexed and unprocessed text
particular feature, in which potential customers are that reflects the opinions of people. Many people make
interested. The authors conduct experiments on the sentiment choices by taking the suggestions of other people into
mining and retrieval system using the customer reviews of account. For example, one likes to buy a product that is most
four kinds of electronic products including digital cameras, recommended by people who use that product. Thus, there is
cell phones, laptops, and MP3 players. The evaluation results a need to crawl and process peoples’ opinions, so that it can
indicate that the proposed approach achieves a precision of be used in decision making processes of potential Web
%85 approximately. review applications.
Reference [9] describes a multi-knowledge based In this study, we propose a blog mining system that will
approach that utilizes WordNet for statistical analysis and extract movie comments from Web blogs and that will show
movie knowledge. WordNet is a large lexical database of Web blog users what other people think about a particular
English, developed under the direction of George A. Miller movie. Figure 1 shows the overall process model of the
[10]. Nouns, verbs, adjectives, and adverbs are grouped into proposed system. The blog mining process consists of
sets of cognitive synonyms, each expressing a distinct following three main steps: Web crawling, sentiment
concept. The proposed approach, described in [9], analysis, and visualization.
decomposes the problem of review mining and Web crawling: A Web crawler (also known as a Web
summarization into the following subtasks: identifying spider, or Web robot) is a program or automated script that

78

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
browses the World Wide Web in a methodical, automated Visualization: We utilize the Zed Graph [14] for
manner. A Web crawler is a type of software agent that takes visualization to present our findings. Zed Graph provides an
a list of URLs, called seeds, to visit as input. As the crawler ASP web-accessible control for creating 2D line, bar, and pie
visits these URLs, it identifies all the hyperlinks in the page graphs of arbitrary datasets. It is maintained as an open-
and adds them to a list of URLs, called the crawl frontier, to source development project. We presented the results on the
visit. URLs from the frontier are recursively visited project Web site over a shared database.
according to a set of policies. The process of Web crawling
is also known as spidering. Many sites, search engines in
particular, use spidering as a means of providing up-to-date IV. SYSTEM ARCHITECTURE
data. Web crawlers (or spiders) are mainly used to create a
copy of all the visited pages for later processing by a search The proposed system architecture consists of several
engine that will index the downloaded pages to provide fast components: Blog Crawler, Sentiment Analyzer, and Web
searches. Crawlers can also be used for automating Usage Interfaces.
maintenance tasks on a Web site, such as checking links or
validating HTML code. Also, crawlers can be used to gather A. Blog Crawler
specific types of information from Web pages, such as One of the most important parts of the proposed system
harvesting e-mail addresses (usually for spam) or gathering is the Blog crawler. The crawler needs to analyze as much
text content. In this study, we utilized two open source data as possible to provide good accuracy results. If the
projects, OpenWebSpider [15] and Arachnode [20], for analysis has not been conducted with enough data, the results
crawling the Web blogs and collecting data for sentiment will only indicate opinions of restricted group of people.
analysis. Although one needs to crawl as many blogs as it is possible
to reach good results, the blogosphere contains huge amounts
of data. The storage capacity is limited and there also exist
limitations related to the computation and memory
capabilities to crawl all of the blogosphere. Thus, in this
study, to calculate the general opinions of people about a
movie, we were only able to crawl some part of the
blogosphere. We can assume that when the computation
capabilities are improved and crawled part of the
blogosphere is increased, the proposed application will
produce better results.
We use Arachnode.Net to crawl the Web blogs.
Arachnode.net is an open source Web crawler for
downloading, indexing, and storing Internet content
including e-mail addresses, files, hyperlinks, images, and
Web pages. Arachnode.net is written in C# and uses SQL
Server 2005. Arahnode.net uses the Lucene.Net library for
indexing and searching. Arachnode.Net is selected because it
is very customizable and well written. We start crawling with
seeds like www.blogpulse.com and www.technorati.com,
Figure 1 Blog Miner Overall Process Model
because these Web sites contain many links to Web blogs. In
turn, this improves the crawling performance. Figure 3
Sentiment analysis: Sentiment analysis has three main
shows the main working process of the crawler.
tasks: determining subjectivity, determining sentiment
orientation, and determining the strength of the sentiment
orientation. In this study, we use an unsupervised approach
in sentiment analysis. We use OPEN-NLP to find types of
words. In our approach, we use a keyword database, which
contains specific words about a movie domain. We use a
keyword algorithm, which searches the keywords in the text.
If a keyword under consideration is found in the database,
then the algorithm identifies whether the keyword is an
adjective or an adverb and calculates the score. An
alternative algorithm, the all-word algorithm, is to look at all
words in related sentences and then calculate the general
score for a movie.

79

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
 Figure 2 Web Crawler Architecure Figure 3 SentiWord Data Table

If the analyzer finds a pre-defined keyword in a sentence

B. Sentiment Analyzer of a given blog page for a specific movie, it looks for the
The Sentiment analyzer is a crucial component of the sentiment words (such as an adjective or an adverb) that
proposed system. It calculates the sentiment scores for a may be associated with that keyword. If it finds a sentiment
movie for different keywords by mining the comments from word, it obtains its score from SentiWord. It uses the
blog pages. In order to calculate the sentiment scores, first, obtained score as the keyword’s score and adds that to the
the analyzer selects the blog pages that contain comments total sentiment score of the blog page.
about a specific movie. Then, it parses each blog text and Say, if the analyzer finds an adjective in a sentence of a
processes it in order to calculate sentiment scores for given blog for a specific movie, it also looks for an adverb
different keywords related to the movie under consideration. that modifies the degree of the adjective. Here, the adverbs
The Sentiment analyzer utilizes the aforementioned are separated into two main categories, degree-adverbs and
keyword algorithm in order to calculate sentiment scores. In reversing-adverbs. If the analyzer finds a degree-adverb
this algorithm, the sentiment analyzer processes every such as “less” or “more” in front of an adjective, then it
sentence of a blog page for the keywords, such as “Screen multiplies the adjective’s score with the degree-adverb’s
Play”, “Director”, and “Producer”, that are related to the score and uses the result as the keyword’s score. If the
movie domain. analyzer finds a reversing-adverb such as “not” in front of
The analyzer utilizes the SentiWordNet [11] to obtain an adjective, it simply reverses the score of that adjective
the sentiment scores. The SentiWordNet is a lexical and uses the result as keyword’s score.
resource, where each WordNet [10] synset s is associated to This way, the analyzer calculates the accumulative
three numerical scores Obj(s), Pos(s) and Neg(s), describing sentiment scores for all related blog pages for different pre-
how objective, positive, and negative the terms contained in defined keywords and takes the average of these scores. In
the synsets are. Figure 3 shows some adjectives and their the end, the analyzer finds a sentiment score corresponding
scores according to the SentiWordNet. to each pre-defined keyword by mining the blogs for each
particular movie.
The peoples’ reviews and comments in their blogs may
contain spelling errors and these errors will decrease
accuracy of the application. To overcome this challenge
NetSpell [15] is used as a spell checker library in our score
calculation methodology.

80

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
 Figure 5 shows the Entity Relationship Diagram of the
proposed application. Note that this diagram does not
include the Arachnode.Net database, which is used to store
blog pages. The database diagram of the Arachnode.Net is
available at [12]. In Figure 5, the “Movies” table is used to
store the score results of each movie under investigation.
The “People” table is used to store all related information
about people who play a role in movies such as actor,
actress, and director. The “SentiWord” table stores the
sentiment dictionary, which was obtained from
SentiWordNet [11]. The “Movie Elements” table stores the
9 keyword categories from the movie domain, and the
“Element Alias” table stores the keywords associated to
these categories.

C. Web User Interfaces

We developed a Web user interface, as illustrated in
Figure 4 Sentiment Analyzer Process Model Figure 6, to present the project evaluation and to give
information about ongoing research. The Web user interface
The SentiWordNet database contains the stem of the is formed mainly under two categories. The first category is
words. In turn, this may affect the calculation of the the selection. There are two types of selection options. First
sentiment scores and decrease accuracy of the application. is the selection of movies. Here, the system lets the user
Thus, in order to find out the sentiment score of a word, the select a movie and then shows the sentiment score results
analyzer must search its stem within the SentiWordNet. To corresponding to nine different keyword categories. Second
overcome this problem, the analyzer utilizes the Porter is the selection of keyword categories. Here, the system lets
Stemmer [13] to get the stem of a word. These text and the user specify only one category and shows the sentiment
word modifications improve the proposed application’s scores of different movies under the selected keyword
accuracy. Figure 4 shows the process model of the category. In addition, the system also lets users select the
sentiment analyzer. movies they want to sketch and the category under which
they want to do the analysis and shows the results in a graph.

Figure 5 Blog Miner ER Diagram Figure 6 The graphs page of the Application

81

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
 The second category is the graphs. The system utilizes We present our experimental study by showing the steps
dynamic charts that are created each time users specify a of the BlogMiner application for processing the raw data and
selection as illustrated in Figure 7. Here, we utilize Zed calculating sentiment scores. Thus, the following sample
Graph [14], which is an open-source library, written in C#, user-review is chosen from IMDB to illustrate the steps.
for creating 2D line and bar graphs of arbitrary datasets. This
library provides a high degree of flexibility, i.e., almost Sample Review: “I thought it wouldn't be as good as it
every aspect of the graph can be user-modified. Zed Graphs was, because thousands of people and reviews said it would
has two different libraries that can be used for creating suck! It was great, but what it missed was that it needed to be
Windows-based applications and Web-based applications. In at-least an hour longer, because it missed a-little bit, but it
this study, we use only some parts of the Zed Graph libraries still rocked! I loved it! I thought it was funny, and as did the
to create a Web-based BlogMiner application. person next to me, when John says: "I'll be back!””.

Step 1-Split the text into sentences: In this step, the

BlogMiner simply takes the text apart into sentences and
makes the sentiment analysis at sentence level. Below, we
illustrate the sample review after step 1.

~1~ I thought it wouldn’t be as good as it was, because thousands of

people and reviews said it would suck! ~1~
~2~ It was great, but what it missed was that it needed to be at-least
an hour longer, because it missed a-little bit, but it still rocked! ~2~
~3~ I loved it! ~3~
~4~ I thought it was funny, and as did the person next to me, when
John says: "I'll be back!””. ~4~

Step 2-Tag the words in each sentence by their type: In

Figure 7 Web User Interface Process Model this step, appropriate tags are added to the words to be able
to understand their meanings more accurately. Figure 8
shows the tags that have been used and the meanings of these
V. EXPERIMENTS AND RESULTS
tags. The text below is the sample review after step 2.

To evaluate the performance of the proposed application,

we use reviews about few movies from The Internet Movie I/PRP thought/VBD it/PRP would/MD not/RB be/VB as/RB
Database (IMDD) Web site [17] as the data set. For our good/JJ as/IN it/PRP was/VBD ,/, because/IN thousands/NNS of/IN
analysis, we simply choose recent movies, since we want to people/NNS and/CC reviews/NNS said/VBD it/PRP would/MD
analyze as many user comments as possible. Our assumption suck/VB !/.
is that recent movies will attract more user comments, as
It/PRP was/VBD great/JJ ,/, but/CC what/WP it/PRP missed/VBD
they may have a larger audience. Thus, we chose following
was/VBD that/IN it/PRP needed/VBD to/TO be/VB at-least/JJ
10 movies from the IMDB: “The Fast and Furious”, an/DT hour/NN longer/RB ,/, because/IN it/PRP missed/VBD a-
“Monsters vs. Aliens”, “State of Play”, “Knowing”, “The little/JJ bit/NN ,/, but/CC it/PRP still/RB rocked/VBD !/.
Dark Knight”, “Wall-E”, “Slumdog millionaire”, “No
Country for old men”, “There Will be Blood” and “The I/PRP loved/VBD it/PRP !/.
Curious Case of Benjamin Button”. For each movie,
approximately 10 review pages are crawled by the Blog I/PRP thought/VBD it/PRP was/VBD funny/JJ, /, and/CC as/RB
Crawler. In turn, this created approximately 1000 user di/VBD the/DT person/NN next/JJ to/TO me/PRP, /, when/WRB
reviews in total. These reviews are used for experiments to John/NNP says/VBZ :/: "/`` I/PRP will/MD be/VB back/RB !/.
calculate the accuracy of application. For the pre-defined ”/NN. /.
keywords, we use the names of the three most important
roles for each movie: actor/actress, director, and Step 3-Point the text using the keyword algorithm and
screenwriter. We include the names of these roles in the calculate the scores: In this step, the system calculates the
database in order to catch comments about actors, actresses, sentiment score for keywords and finds the accumulated
directors, and screenwriters. We refer the readers to scores for each sentence. Below, we illustrate the output of
reference [18] for extensive discussion on implementation the sample review after step 3.
and experiments.

82

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
 “I/PRP thought/VBD it/PRP would/MD not/RB<-1> be/VB as/RB
good/JJ<0.844> as/IN it/PRP was/VBD ,/, because/IN
thousands/NNS of/IN people/NNS and/CC reviews/NNS said/VBD
it/PRP would/MD suck/VB !/.
(sentence score = -0.844)

It/PRP was/VBD great/JJ<0.344> ,/, but/CC what/WP it/PRP

missed/VBD was/VBD that/IN it/PRP needed/VBD<-0.140625>
to/TO be/VB at-least/JJ an/DT hour/NN longer/RB ,/, because/IN
it/PRP missed/VBD a-little/JJ bit/NN ,/, but/CC it/PRP still/RB<-
0.171> rocked/VBD !/.
(sentence score = 0.0104)

I/PRP loved/VBD<0.375> it/PRP !/.

(sentence score = 0.375)

I/PRP thought/VBD it/PRP was/VBD funny/JJ<-0.515> ,/, and/CC

as/RB did/VBD the/DT person/NN next/JJ to/TO me/PRP ,/, Figure 9 Experiment Results
when/WRB John/NNP says/VBZ :/: "/`` I/PRP will/MD be/VB
back/RB !/. ””/NN ./. The results of the experiment have been compared with
(sentence score = -0.515) each movie’s IMDB score. On the IMDB page of each
movie, the movie’s general scores are listed. Thus, we can
compare the IMDB score against the keyword algorithm’s
The results of the experiments are illustrated in figure 9 score.
below. As it can be seen in the figure, in producer and screen For producer and screen writer categories not enough
writer columns, there are rows that have a score of 5.25. comments were found to calculate a realistic score. Because
These scores are default values because no keywords were of this, most of the producer and screenwriter score columns
found for these movies. are the default value. When the results are compared against
the IMDB scores, we observe a similar behavior. A movie
with a low IMDB score also gets a low score in the proposed
application. Similarly, a movie with high IMDB score gets a
high score in the proposed application. We also observe two
exceptions to this behavior. For example, the movies “Fast
and Furious” and “State of Play” are getting a high scores,
however, their IMDB scores are in a lower position than the
proposed application calculated. We conclude that in the
IMDB database, the comments and the score of the movie
may not always have a correct match.

VI. CONCLUSION AND FUTURE RESEARCH DIRECTIONS

Opinion mining is an important area of investigation. As

the Web 2.0 applications produce enormously and collecting
meaningful information, mining such information has
become an important task. In this study, we introduced an
opinion mining application that is created for calculating
movie scores from Web blog pages.
Figure 8 Word Tags Experiment results show that the proposed application
produces accurate results close to real values. With this
study, we introduced an unsupervised approach for sentiment
analysis. For the future study, we want to further improve
this application and investigate how supervised learning
methodologies can further improve the accuracy in the
results.

83

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.
Acknowledgement: We thank Kadir Ardic and Onur Enez for [13] Web site for Porter Stemmer is available at
their contribution to the research presented in this paper. We http://tartarus.org-/~martin/PorterStemmer, Access
also thank the Department of Computer Engineering in data: October 2009
Marmara University for giving us permission to commence [14] Web site for Zed Graphs is available at
this study and to do the necessary research work by utilizing http://zedgraph.org-/wiki/index.php?title=Main_Page,
departmental computer facilities. Access date: October 2009
[15] Web site for NetSpell is available at
http://sourceforge.net-/projects/netspell, Access date:
REFERENCES October 2009
[16] Web site for OpenWebSpider is available at
[1] Bing Liu, Web Data Mining - Exploring Hyperlinks, http://www.openwebspider.org, Access date: October
Contents and Usage Data, Text Book, , Springer, 2009
December, 2006 [17] Web site for The Internet Movie Database (IMDB) is
available at http://www.imdb.com, Access date:
[2] Technorati Web Site is available at
http://technorati.com, Access Data: October 2009 October 2009
[3] Zhongchao Fei, et al., Sentiment Classification Using [18] Kadir Ardic, Onur Enez, Undergraduate graduation
thesis. available at http://www.scribd.com/doc/-
Phrase Patterns Proceedings of the Fourth International
16191423/Web-Blog-Miner-Licence-Thesis, Access
Conference on Computer and Information Technology
date: October 2009
(CIT’04), 2004.
[4] Jeonghee Yi, et al., Sentiment Mining in WebFountain,
Proceedings of the 21st International Conference on
Data Engineering (ICDE 2005), 2005
[5] Jian Liu, et al., Super Parsing: Sentiment Classification
with Review Extraction, Proceedings of the Fifth
International Conference on Computer and Information
Technology (CIT’05), 2005.
[6] Yun-Qing Xia, et al., The Unified collocation
Framework for Opinion Mining, Proceedings of the
Sixth International Conference on Machine Learning
and Cybernetics, Hong Kong, 19-22 August 2007.
[7] Qingliang Miao, et al., AMAZING: A sentiment mining
and retrieval system, Expert Systems with Applications
(2008) doi:10.1016/j.eswa.2008.09.035.
[8] Qiang Ye, et al., Sentiment classification of online
reviews to travel destinations by supervised machine
learning approaches, Expert Systems with Applications
(2008) doi:10.1016/j.eswa.2008.07.035.
[9] Li Zhuang, et al., Movie review mining and
summarization, Proceedings of the 15th ACM
international conference on Information and knowledge
management, 2006.
[10] WordNet Web site is available at
http://wordnet.princeton.edu, Access Date: October
2009.
[11] Andrea Esuli, et al., SENTIWORDNET: A Publicly
Available Lexical Resource for Opinion Mining, The
fifth international conference on Language Resources
and Evaluation, LREC 2006
[12] Web site for Arachnode.Net Database Diagrams is
available at http://arachnode.net/media/g/database_-
diagrams/default.aspx, Access date: October, 2009

84

Authorized licensed use limited to: muthu m. Downloaded on June 15,2010 at 06:11:02 UTC from IEEE Xplore. Restrictions apply.