JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 68

Unsupervised Approaches for Detection of

Web Users Opinion on Products
Anil Kumar.K.M, Anil Kumar.P and Suresha

Abstract— Opinion mining is a recent sub discipline of information retrieval which is not about the topic of a document, but with the
opinion it expresses. There are number of research activities that focused on determination of the semantic orientation at the word
level, since seed words play an important role as sentiment clues in indicating the sentiment of a sentence or a document. Web
users document their opinion in the form of reviews or opinionated texts at different opinionated sites, shopping sites, personal pag-
es etc., to express and share their opinion with other web users. The opinion expressed by web users may be on diverse topics
such as politics, sports, products, movies etc. These opinions will be very useful to others such as, leaders of political parties, selec-
tion committees of various sports, business analysts and other stake holders of products, directors and producers of movies as well
as to the other concerned web users. In this paper, we present an unsupervised sentence and document based approaches for find-
ing opinion of web users from opinionated texts and to classify web user’s opinion into positive or negative. Our approach performs
better when subjected to different data sets of nearly four hundred and forty five opinionated texts. The results of our approach are
good compared to other published results.
Index Terms— Opinion Detection, Product Analysis, Phrase Detection, Sentiment Analysis.

——————————  ——————————

1 INTRODUCTION

W
ith the rapid expansion of e-commerce, more and
more products are sold on the Web, and more
and more people are also buying products online.
In order to enhance customer satisfaction and shopping
experience, it has become a common practice for online
merchants to enable their customers to review or to ex-
press opinions on the products that they have purchased.
With more and more common users becoming comforta-
ble with the Web, an increasing number of people are
writing reviews. As a result, the number of reviews that a
product receives grows rapidly. Some popular products
can get hundreds of reviews at some large merchant sites.
Furthermore, many reviews are long and have only a few
sentences containing opinions on the product. This makes
it hard for a potential customer to read them to make an
informed decision on whether to purchase the product. If
customer reads only few reviews, he/she may get a bi-
ased view. The large number of reviews also makes it
hard for product manufacturers to keep track of customer
opinions of their products. For a product manufacturer,
there are additional difficulties because many merchant
sites may sell its products, and the manufacturer may
(almost always) produce many kinds of products.[2]
The development of web and its related technologies
have fueled the popularity of the web with all sections of
society. The web has been rightfully used by govern-
ments, business houses, industries, educational institu-
————————————————
 Anil Kumar.K.M is with Sri Jayachamarajendra College of Engineering,
Mysore, Karnataka, India.
 Anil Kumar.P is with Sri Jayachamarajendra College of Engineering,
Mysore,Karnataka, India.
 Suresha is with Dept of studies in Computer Science, University of My-
sore.
tions etc., to make them available globally and also gain
knowledge globally. The individual user, on the other
end, is provided with an opportunity to gain knowledge
and to share knowledge. The web is the source of many
research activities and one interesting area is to mine user
opinion from web on diverse topic. The study of opinions
is useful for both producers and consumers of the topic.
The producers can be manufacturers of digital products,
automobile manufactures, movie producers, editor of
news article etc., very much interested to find opinion of a
user. The consumers are individual user’s who document
their opinion and want to share it with others about the
topic [14].
Many readers of online reviews say that these reviews or
opinions influence their purchasing decision [4]. Today
people of all ages and from all over the world use web for
collecting opinions. There are many sites which allow
user’s to express their opinion such as Epinions, Amazon,
CNet[5][6]and [7] etc., some of these sites are supervised
manually, therefore all opinions expressed by user’s may
not be published. The opinionated sites seem to often ex-
hibit highly skewed rating distributions with a particular
bias towards positive reviews [8][9][10]and [11]. There are
cases in opinionated sites where the user’s have selected a
low rating while their opinion indicates a positive opinion
[12].
Before the Web, when an individual needs to make a de-
cision, he/she typically asks for opinions from friends
and families. When an organization needs to find opi-
nions of the general public about its products and servic-
es, it conducts surveys and focused groups. With the
Web, especially with the explosive growth of the user

© 2010 Journal of Computing Press, NY, USA, ISSN 2151-9617

http://sites.google.com/site/journalofcomputing/
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
generated content on the Web, the world has changed.
One can post reviews of products at merchant sites and
express views on almost anything in Internet forums, dis-
cussion groups, and blogs, which are collectively called
the user generated content. Now if one wants to buy a
product, it is no longer necessary to ask one’s friends and
families because there are plentiful of product reviews on
the Web which give the opinions of the existing users of
the product. For a company, it may no longer need to
conduct surveys, to organize focused groups or to employ
external consultants in order to find consumer opinions
or sentiments about its products and those of its competi-
tors. Finding opinion sources and monitoring them on the
Web, however, can still be a formidable task because a
large number of diverse sources exist on the Web and
each source also contains a huge volume of information.
In this paper we focus on detecting opinions expressed by
web users using both Sentence and Document based ap-
proaches and provide a summary of users opinion. The
remainder of this paper is organized as follows: In Section
2 we give a brief description of related work. Then, in
Section 3, we discuss our methodology. In Section 4, the
experimental results are discussed. We discuss applica-
tion in section 5. Conclusion is discussed in Section 6.
2 RELATED WORK
Opinion mining is a recent sub discipline of information
retrieval which is not about the topic of a document, but
with the opinion it expresses [28][29]. In literature opi-
nion mining is also known as sentiment analysis[30], sen-
timent classification [31], opinion extraction[32], affective
classification[29] and affective rating[33]. It has emerged
in the last few years as a research area, largely driven by
interests in developing applications such as mining opi-
nions in online corpora, or customer relationship man-
agement e.g., customer’s review analysis[29].
There are number of research activities that focused on
determination of the semantic orientation at the word
level, since seed words play an important role as senti-
ment clues in indicating the sentiment of a sentence or a
document. Hatzivassiloglou and McKeown [34] have at-
tempted to predict semantic orientation of adjectives by
analyzing pairs of adjectives (i.e., adjective pair is adjec-
tives conjoined by and, or, but, either-or, neither-nor) ex-
tracted from a large unlabelled document set.
Turney has obtained remarkable results on the sentiment
classification of terms by considering the algebraic sum of
the orientations of terms as representative of the orienta-
tion of the document. Turney and Littman have boot-
strapped from a seed set, containing seven positive and
seven negative words, and determined semantic orienta-
tion according to Point wise Mutual Information-
Information Retrieval (PMI-IR) method.
69
Wang and Araki [35] proposed a variation of the Seman-
tic Orientation-PMI algorithm for Japanese for mining
opinion in weblogs. They applied Turney method to Jap-
anese webpage and found results slanting heavily to-
wards positive opinion. They proposed balancing factor
and neutral expression detection method and report a
well balanced result.
Kamps et al [16] have focused on the use of lexical rela-
tions, defined in Word Net. They defined a graph on the
adjectives contained in the intersection between the Tur-
ney’s seed set and Word Net, adding a link between two
adjectives whenever WordNet indicate the presence of a
synonymy relation between them. The author’s defined a
distance measure d (t1, t2) between terms t1 and t2, which
amounts to the length of the shortest path that connects t1
and t2. The orientation of a term is then determined by its
relative distance from the seed terms good and bad.
Esuli and Sebastiani [17] proposed semi-supervised learn-
ing method started from expanding an initial seed set
based on Turney and Littman’s seed set [18], by using
WordNet. Their basic assumption is terms with similar
orientation tend to have similar glosses. They determined
the expanded seed term’s semantic orientation through
gloss classification by statistical technique.
Morinaga et al [19] presents a new frame work where a
user inputs products names and the system collects
people’s opinion and attaches three labels to each, such as
the name of the product, positive or negative nature of
opinion and opinion likeliness. These labeled opinions are
then put into an opinion database. A user specifies a tar-
get category for analysis using the value of labels. The
system conducts text mining to extract statistically mea-
ningful information corresponding to the specified target.
Pang et al [20] adopted a statistical technique-based ap-
proach, using supervised machine learning with words
and n-grams as features to predict orientation at the doc-
ument level.
Kim and Hovy [21] presents orientation detection system
that assigns to each term, a positive score and a negative
score, the terms may have both a positive and a negative
correlation, with different degrees, and some terms may
carry a stronger positive or negative orientation than oth-
ers. Their system starts from a set of positive and negative
seed terms, and expands the positive and negative seed
set by adding to it the synonyms of positive and negative
seed terms and the antonyms of negative and positive
seed terms. The system classifies then a target term t into
either positive or negative by means of two alternative
learning-free methods based on the probabilities that
synonyms of t also appear in the respective expanded
seed sets.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
Popescu and Etzioni [22] introduced OPINE. It is an un-
supervised information extraction system that outputs set
of features which is accompanied by a list of opinions that
are ranked based on strength for a given product and
corresponding reviews. To find features, it first extracts
the noun phrases from reviews and retains those with
frequency greater than an experimentally set threshold.
OPINE’s Feature Assessor evaluates each noun phrase by
computing the PMI scores between the phrase and mero-
nymy discriminators associated with the product. It ex-
tracts opinion phrases, which are adjective, noun, verb or
adverb phrases representing customer opinions and uses
relaxation labeling, unsupervised classification technique,
for finding the semantic orientation of words.
Pang and Lee [23] report of work in progress on using
simple statistics in an unsupervised fashion to re-rank
search engine results for a review oriented query. They
report that their proposed technique performs compara-
bly to methods that rely on sophisticated pre-encoded
linguistic knowledge.
ReviewSeer is a tool that automates the work done by
aggregation sites. It uses various methods such as meta-
data and statistical substitutions, linguistic substitutions,
language based modifications, n-gram and proximity for
feature extraction. Naive Bayes classifier is used with pos-
itive and negative review sets for assigning a score to the
extracted feature terms. The classifier performed well for
reviews collected from CNet and Amazon for training
and testing. The classifier did not perform well for web
pages crawled from the result of a search engine. It dis-
plays attributes and score of the attribute along with re-
view sentences [24].
WebFountain uses beginning definite Base Noun Phrase
(bBNP) heuristic for extracting product features. It ex-
tracts the base noun phrases at the beginning of the sen-
tences followed by a verb phrases. To assign sentiments
to the features, reviews are parsed and traversed with
two linguistic resources namely the sentiment lexicon and
the sentiment pattern database. The sentiment lexicon
defines the polarity of terms and sentiment pattern data-
base defines sentiment extraction patterns for a sentence
predicate [25].
Red Opal is a tool that assumes online shoppers are high-
ly task driven and have certain goal in mind and that they
are looking for product with features that are consistent
with that goal. This system enables users to find products
based on features. It scores each product based on fea-
tures from the customer reviews. It uses frequent nouns
and noun phrases for feature extraction and user ratings
are used to compute product score for features mentioned
in reviews. The results are shown in descending order for
each feature along with the URL [26].
Opinion observer is a sentiment analysis system for ana-
lyzing and comparing opinions on the web. The product
70
features are extracted from noun or noun phrases by the
association miner. They use adjectives as opinion words
and assign prior polarity of these by WordNet exploring
method. The polarity of an opinion expression which is a
sentence containing one or more feature terms and one or
more opinion words is assigned a dominant orientation.
The extracted features are stored in a database in the form
of feature, number of positive expression and number of
negative expression. The system shows the results in a
graph format showing opinion of the product feature by
feature [27].
Previous works on mining opinions can be divided into
two directions: sentiment classification and sentiment
related information extraction. The former is a task of
identifying positive and negative sentiments from a text
which can be a passage, a sentence, a phrase and even a
word. The latter focuses on extracting the elements com-
posing a sentiment text. The elements include source of
opinions who expresses an opinion. Some researchers
refer this information extraction task as opinion extraction
or opinion mining. Comparing with the former one, opi-
nion mining usually produces richer information [24].
3 METHODOLOGY
In this paper, we present unsupervised approaches to
detect opinion of web users from product reviews and to
classify web user’s opinion into positive or negative.
People use phrases to express their opinions to a consi-
derable extent, based on these phrases, the opinions are
detected using two different approaches (viz. Sentence
based and Document based) and classify them as positive
or negative. The system will detect the opinion and con-
vey whether the opinion of public about the entity is posi-
tive or negative. The system is tested against nearly four
hundred and forty five files and the accuracy of the result
is encouraging and good.
3.1 Sentence based Approach
The Sentence based approach contains five steps. In the
first step, the file containing product reviews is split into
sentences and then the review text is tagged using Monty
tagger. The tagged text is used for pattern detection for
user defined patterns. The phrases are extracted based on
the patterns and each phrase is checked to obtain the
polarity. Based on the window size, the polarity of each
phrase is calculated authenticating the presence in seed
lists and total score of the review is calculated. The flow
chart for the sentence based approach is shown in Fig.1.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
File Containing Review
Split review into sentences
Apply Monty tagger
Phrase Detection Check for
Polarity
Opinion Detection
Fig.1 Flow chart of design process for sentence approach
 File containing review- In this step, the product re-
views are given as input to the application. The en-
tered review is stored in a unique text file and the file
names are given in serial order. These files are given
as input to the sentence split module.
 Split review into Sentences - In this step, the prod-
uct reviews are split into sentences assuming “.”as
the delimiter for all the sentences in the review. The
algorithm checks for the presence of “.” (Dot). If “.” is
found, then a new line is inserted. This will be re-
peated for the entire review and the output is written
into a new file.
 Parts of speech tagger - In this step, we apply the
Monty tagger [31] to the original review file which is
already split into sentences to obtain the tagged text
file. The need of this step is to obtain the sentences
with part of speech tags that help to identify phrases
which are adjectives.
 Phrase detection – In this step, the output from the
previous text, i.e. the tagged review is taken as input.
The review will be in the form of sentences with
tagged text. The phrases are detected based on the
matching adjectives. For this, all the words of a sen-
tence are stored in an array and each word is checked
against the pattern. We search for the phrases that
were found from the tagged reviews in the seed lists.
If the phrase is found in positive seed list, the sen-
tence in which the phrase was found is assigned +1
and if it is found in negative seed list, the phrase is
assigned -1. The score for each sentence is stored for
further calculation.
 Check for Polarity – In this step, the output from the
phrase detection is taken as input. A threshold with
window size five is selected for checking the polarity.
Each identified phrase is checked within five words
to the left and five words to the right of the sentence.
If “not” is found within the window size either to the
left or right, the polarity will be negated by multiply-
71
ing with -1. The score will be calculated by adding
the left count and right count of the window size for
each identified phrase and total score of the sentence
is calculated.
 Opinion Detection – Total score for the review file is
calculated by adding the scores obtained by pattern
detection step and check for polarity step for each
sentence. The total score of the review is calculated as
the sum of scores of all the sentences in the review
file. We use here 0 as threshold. If the score of the re-
view is greater than threshold T, the opinion is consi-
dered as positive and if the score is less than thre-
shold, then the opinion is considered as negative.
3.2 Document based Approach
The Document based approach contains four steps. In the
first step, the file containing review is tagged using
Monty tagger. The tagged text is used for feature
extraction for user defined patterns. The phrases are
extracted based on the patterns and each phrase is
checked for polarity. The polarity of each phrase is
calculated by authenticating its presence in positive and
negative list. The total score of the review is calculated by
adding the polarity of all the features extracted. The flow
chart for the document based approach is shown in Fig.2.
File Containing Review
Apply Monty tagger
Check for
Feature Extraction Polarity
Opinion Detection
Fig.2 Flow chart of design process for Document based approach
The algorithm extracts phrases containing adjectives or
adverbs. This is because the research has shown that the
adjectives and adverbs are good indicators of subjectivity
and opinions. However, although an isolated adjective
may indicate subjectivity, there may be an insufficient
context to determine its opinion orientation. Therefore,
the algorithm extracts two consecutive words, where one
member of the pair is an adjective/adverb and the other
is a context word. Two consecutive words are extracted if
their Part Of Speech (POS) tags conform to any of the pat-
terns in Table 1.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 72

4 EXPERIMENTS AND RESULTS

TABLE 1: ORDER OF PATTERNS CONSIDERED FOR EXTRACTION
We conducted many experiments with our proposed
OF PHRASES
approaches. The details of the experiments and
First word Second word Third word
corresponding results are discussed below
(not extracted)
JJ/JJR/JJS NN/NNS/NNP Anything
/NNPS Experiment - 1: Phrase detection using Sentence based
JJ/JJR/JJS JJ/JJR/JJS not NN nor Approach
NNS
RB/RBR/RBS/RBP JJ/JJR/JJS not NN nor Example – 1:
NNS
NN/NNS/NNP/N JJ/JJR/JJS not NN or NNS
NPS In this experiment, the tagged review file split into
RB/RBR/RBS/RBP VB/VBD/VBN/ anything sentences is given as input. The output is the score of the
VBG/VBP/VBZ review for all sentences using Sentence based Approach.

 Check for Polarity – This step deals with checking Result-1: The output of the file for the input
the polarity of the features extracted in the previous (inp/04_nokia6610.txt) is written into a new file. The Score
step. The output from the features extraction is taken obtained is greater than threshold (T > 0) and hence opi-
as input. The phrases extracted are analyzed for their nion is classified as Positive.
polarity. The extracted phrases contain adjectives or
adverbs. The extracted phrases are searched for their Generated Report - Sentence Approach
presence in the positive list and negative list. If the -------------------
Inp/04_nokia6610.txt Score=6 Opinion : Positive
feature is present in the positive list, a score of +1 will
be assigned. Similarly, if the feature is present in the
negative list, then a score of -1 will be assigned. This Example – 2
process is applied for all the features extracted from
the document. Result -2: The output of the file for the input
(inp/16_Canon.txt ) is written into a new file as shown below.
 Opinion Detection – Once the polarity of the doc- The Score obtained is greater than threshold (T > 0) and
uments is known, the final step deals with the analy- hence opinion is classified as Positive.
sis of the Statistics. These statistics are used to deter-
mine the quality of the product. First the documents Generated Report - Sentence Approach
which feel positive are passed through entire process -------------------
and result of that is calculated in terms of percentage
Inp/16_Canon.txt Score=5 Opinion : Positive
later negative set of documents is passed through the
process. The statistics of the entire document is taken.
The score of a document review is calculated by add-
Experiment - 2: Feature extraction using Document
ing the positive score and negative score. We use
based Approach
here 0 as threshold T. If the score of the review is
greater than threshold T, the opinion is considered as
Example – 1
positive and if the score is less than threshold, then
the opinion is considered as negative.
In this experiment, the tagged review file is given as in-
put. The output is the score of the review using Docu-
ment based Approach.
The product with a higher percentage of positive
results yields in good product and the products with Result -1: The output of the file for the input
higher percentage of negative results will end up with a (inp/04_nokia6610.txt) is written into a new file. The Score
bad product. The reviews help the manufacturers in obtained is greater than threshold (T > 0) and hence opi-
knowing the positive and negative features of their nion is classified as Positive.
product and thus further help them in developing better
Inp/04_nokia6610.txt – Document Approach
products. There is a huge need in the industry for such
Inp/04_nokia6610.txt Score=6 Opinion : Positive
services because every company wants to know how
consumers perceive their products and services and those
of their competitors.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 73

Example – 2
TABLE: 3 CLASSIFICATION OF MULTIPLE REVIEWS USING
Result – 2: The output of the file for the input DOCUMENT BASED APPROACH ON DATA SET 1
(inp/16_Canon.txt) is written into a new file as shown below.
Results obtained using Document based Approach
The Score obtained is less than threshold (T < 0) and
hence opinion is classified as Negative. Sl Total Positive Negative Avg
Product
N No.of Accuracy Accuracy Score
inp/16_Canon.txt – Document Approach Name
o Files (%) (%) (%)

Inp/16_Canon.txt Score:-1 Opinion : Negative 1 NikonCoolPix 33 70.00 100.00 72.73

2 Nokia6610 34 96.88 50.00 94.12

Experiment - 3: Classification of Multiple Reviews 3 Canon 45 59.52 66.67 60.00
(Dataset 1)
4 CreativeZen 95 85.33 90.00 86.32
In this experiment, the proposed approaches are tested 5 Apex_DVD 97 58.46 75.00 63.92
against multiple reviews of product data set 1 and the
detection of opinion in terms of percentages is calculated. We obtain a positive accuracy of nearly 74% and negative
We conducted the accuracy test as shown below. accuracy of nearly 76% for Data set 1 with Document
based approach. Similarly, positive accuracy of nearly
Accuracy = (true positive + true negative) ÷ 57% and negative accuracy of nearly 32% is obtained for
(total number of items to be classified)
Data set 2 using the Document based approach.

Table 2 shows the product wise reviews considered for

classification and their classification as positive or nega-
tive using Sentence based Approach. Table 3 shows the
product wise reviews considered for classification and
their classification as positive or negative using Docu-
ment based Approach. The results obtained using Sen-
tence based and Document based Approaches are good
and encouraging.
Experiment - 4: Classification of Multiple Reviews
(Dataset 2)
In this experiment, we obtained data set containing mul-
tiple reviews from [36]. This data set contained 70 reviews
which were positive and 71 reviews that were truly nega-
tive.

TABLE: 2 CLASSIFICATION OF MULTIPLE REVIEWS USING True positive represent number of opinionated texts clas-
SENTENCE BASED APPROACH ON DATA SET 1 sified correctly as positive, similarly true negative
represent number of opinionated texts classified correctly
Results obtained using Sentence based Approach
as negatives. The result obtained is good considering that
Total Positive Negative Avg we use only adjectives to find opinionated phrases. We
Sl Product
No.of Accuracy Accura- Score obtain an accuracy of 69% and 61% with Sentence based
No Name
Files (%) cy (%) (%)
approach for positive and negative data sets. An accuracy
1 NikonCoolPix 33 93.55 100.00 93.94 of 63% and 63% were obtained with Document based
approach for both positive and negative datasets.
2 Nokia6610 34 96.88 50.00 94.12

3 Canon 45 97.62 33.33 93.33 TABLE: 4 RESULT OF OUR APPROACHES

4 CreativeZen 95 97.33 50.00 87.37
Sl No. of Accuracy
5 Apex_DVD 97 90.77 Approach
53.13 78.35 No reveiws (%)
1 [5] approach 75 84
We obtain a positive accuracy of nearly 95% and negative 2 [5] approach 95 70.53
accuracy of nearly 57% for Data set 1 with Sentence based 3 [5] approach 120 80
approach. Similarly, positive accuracy of nearly 58% and 4 Our Approach 75 91.66
(Sentence Based)
negative accuracy of nearly 33% is obtained for Data set 2
5 Our Approach 95 87.37
using the Sentence based approach.
(Sentence Based)
6 Our Approach 120 87.76
(Sentence Based)
7 Our Approach 75 75.06
(Document Based)
8 Our Approach 95 86.32
(Document Based)
9 Our Approach 120 78.52
(Document Based)
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 74

Table 4 shows result of our approaches against result do-

cumented in literature. Table 5 shows positive accuracy
and negative accuracy of opinionated texts on Data set 1
and Data set 2.

TABLE 5: RESULTS OF TWO APPROACHES

Sl Phrase Detection Ap-
No
proach
1 using Sentence based
approach – Data set1
2 using Sentence based
approach – Data set 2
3 using Document based
approach – Data set 1
4 using Document based
approach – Data set 2
The graph shown in the Figure 3 indicates the positive
and negative accuracy of classification of web users opi-
nion using two proposed approaches.
Fig.3: Accuracy of two approaches on different Data Sets
Data sets Accuracy
Positive 68.90% Fig.5: Accuracy of two approaches on different products : Dataset 2
Negative 60.93%
Positive 62.82% 5 APPLICATION
The field of opinion detection and sentiment analysis is
Negative 62.93%
well-suited to various types of intelligence applications.
Indeed, business intelligence seems to be one of the main
factors behind corporate interest in the field. The task of
collecting opinions and presenting them as a summarized
opinion of the web community will be useful for govern-
ment agencies, industries, educational institutes, financial
sectors etc., which spends time and money in collecting
opinions manually through surveys, polls and other stu-
dies, such studies may not provide the complete view of
the user. Today, many government agencies, industries,
educational institutes, financial sectors etc., are providing
their service on web and would be very interested to find
opinions of their clients. It would be useful for existing
search engines, online markets, review sites etc., for pro-
viding better service to users. It would also benefit indi-
viduals who are interested to find opinion of others on
products. Both, individuals and organizations will benefit
from a summarized opinion resulting in significant sav-
ing of time and money.

The opinion expressed by a web user is detected using 6. CONCLUSION

Opinions are so important that whenever one needs to
adjectives and also other part of speech like verb, adverb
make a decision, one wants to hear others opinions. This
etc. We can obtain better results by considering other part
is true for both individuals and organizations. In this pa-
of speech for detection of web user’s opinion.
per, we have discussed two approaches that detect the
opinion of web users from product reviews that compris-
The Figure 4 shows the average opinion of web users on
es of opinionated phrases using Sentence based approach
different products from data set 1 and Figure 5 shows the
and Document based approaches. Our Sentence based
average opinion of web users on different products from
approach finds opinion using only adjectives. It provides
data set 2 using Sentence and Document based approach-
better results compared to a few published results of oth-
es.
er sentence based approaches. Users express opinions
using adjectives and other part of speech like verb, ad-
verb etc. Our Document based approach finds opinion by
considering adjectives as well as other part of speech. The
result obtained by approach is found to be better than
other document approaches. Finally, our Sentence based
approach is found to be better than our Document based
approach on different Data sets.

Fig.4: Accuracy of two approaches on different products : Dataset 1

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
REFERENCES
[1] Bo Pang and Lillian Lee (2008) "Opinion Mining and
Sentiment Analysis", Foundations and Trends® in In-
formation Retrieval: Vol. 2: No 1–2, pp 1-135.
[2] Minqing Hu and Bing Liu, “Mining and Summariz-
ing Customer Reviews”, Proceedings of KDD’04, Au-
gust 22-25, 2004, Seattle, Washington, USA.
[3] YuanbinWu, Qi Zhang, Xuanjing Huang, LideWu,
“Phrase Dependency Parsing for Opinion Mining”,
Proceedings of the 2009 Conference on Empirical Me-
thods in Natural Language Processing, pages 1533–
1541, Singapore, 6-7 August 2009
[4] ComScore/the Kelsey group, “Online consumer-
generated reviews have significant impact on offline
purchase behavior”, Press Release, November 2007.
www.comscore.com/press/release.asp?press=1928.
[5] Peter D. Turney, Thumbs up or thumbs down? Se-
mantic orientation applied to unsupervised classifica-
tion of reviews”, Proceedings of 40th Annual Meeting
of the Association for Computational Linguistics, pp.
417-424, Philadelphia, 2002.
[6] Kushal Dave, Steve Lawrence and David M. Pennock,
“Mining the peanut gallery: Opinion extraction and
semantic classification of product reviews”, Proceed-
ings of 12th International World Wide Web Confe-
rence, pp. 519-528, Budapest, Hungary, 2003.
[7] Bing Liu, Minqing Hu,“Mining and summarizing
customer reviews”, Proceedings of 10th ACM
SIGKDD International Conference on Knowledge
Discovery and Data Mining, pp. 168- 177, Washing-
ton, 2004.
[8] Judith A. Chevalier and Dina Mayzlin, “The effect of
word of mouth on sales: Online book reviews”, Jour-
nal of Marketing Research, pp. 345-354, 2006.
[9] Michael Dewally, “Internet investment advice: In-
vesting with a rock of salt”, Financial Analysts Jour-
nal, pp. 65-77, 2003.
[10] Daniel Houser and JohnWooders, “Reputation in
auctions: Theory, and evidence from eBay”, Journal
of Economics and Management Strategy, pp. 252-369,
2006.
[11] Nan Hu, Paul A. Pavlou and Jennifer Zhang, “Can
online reviews reveal a products true quality?: empir-
ical findings and analytical modeling of online word-
of-mouth communication”, Proceedings of Electronic
Commerce, pp. 324-330, New York, USA, 2006.
75
[12] Luis Cabral and Ali Hortacsu, “The dynamics of sel-
ler reputation: Theory and evidence from eBay”,
working paper, downloaded version revised in
March, 2006. URL http://pages.stern.nyu.
[13] http://www.cs.uic.edu/~liub/FBS/opinion-
mining.pdf
[14] Anil Kumar K.M and Suresha, “Threshold Based Ap-
proach for Mining Users’ Opinion on Products from
Web” IICAI-2009
[15] Anil, Kumar, K.M., Suresha: Identifying Subjective
Phrases from Opinionated Texts Using Sentiment
Product Lexicon. International Journal of Advanced
Engineering & Applications. 2, 63-271 (2010)
[16] Jaap Kamps, Maarten Marx, Robert J. Mokken and
Maarten De Rijke, “Using wordnet to measure se-
mantic orientation of adjectives”, Proceedings of 4th
International Conference on Language Resources and
Evaluation, pp. 1115-1118, Lisbon, Portugal, 2004.
[17] Andrea Esuli and Fabrizio Sebastiani, “Determining
the semantic orientation of terms through gloss clas-
sification”, Proceedings of 14th ACM International
Conference on Information and Knowledge Man-
agement, pp. 617-624, Bremen, Germany, 2005.
[18] Peter D. Turney and Michael L. Littman, “Measuring
praise and criticism: Inference of semantic orientation
from association”, ACM Transactions on Information
Systems, pp. 315- 346, 2003.
[19] Satoshi Morinaga, Kenji Yamanishi, Kenji Tateishi
and Toshikazu Fukushima, “Mining Product Reputa-
tions on the Web”, Proceedings of 8th ACM SIGKDD
international conference on Knowledge discovery
and data mining SIGKDD’02, pp. 341-349, Edmonton,
Alberto, Canada, 2002.
[20] Bo Pang, Lillian Lee and Shivakumar Vaithyanathan,
“Thumbs up? sentiment classification using machine
learning techniques”, Proceedings of 7th Conference
on Empirical Methods in Natural Language
Processing, pp. 79-86, Philadelphia, US, 2002.
[21] Soo-Min Kim and Eduard Hovy, “Determining the
sentiment of opinions”, Proceedings of 20th Interna-
tional Conference on Computational Linguistics, pp.
1367-1373, Geneva, Switzerland, 2004.
[22] Popescu and Etzioni, “Extracting product features
and opinions from reviews”, Proceedings of Confe-
rence on Human Language Technology and Empiri-
cal Methods in Natural Language Processing, pp.
339-346, Vancouver, British Columbia, Canada, 2005.
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 8, AUGUST 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG
[23] Bo Pang and Lillian Lee, “Using very simple statistics
for review search: An Exploration”, Coling 2008:
Companion volume Posters and Demonstrations, pp.
75-78, Manchester, UK, 2008.
[24] YuanbinWu, Qi Zhang, Xuanjing Huang, LideWu,
“Phrase Dependency Parsing for Opinion Mining”,
Proceedings of the 2009 Conference on Empirical Me-
thods in Natural Language Processing, pages 1533–
1541, Singapore, 6-7 August 2009
[25] Yi and Niblack, “Sentiment Mining in Web Foun-
tain”, Proceedings of 21st international Conference
on Data Engineering, pp. 1073-1083, Washington DC,
2005.
[26] Christopher Scaffidi, Kevin Bierhoff, Eric Chang,
Mikhael Felker, Herman Ng and Chun Jin, “Red
Opal: product-feature scoring from reviews”, Pro-
ceedings of 8th ACM Conference on Electronic
Commerce, pp. 182-191, New York, 2007.
[27] Bing Liu, Minqing Hu and Junsheng Cheng, “Opi-
nion observer: analyzing and comparing opinions on
theWeb”, Proceedings of 14th international Confe-
rence onWorldWideWeb, pp. 342-351, Chiba, Japan,
2005.
[28] Andrea Esuli and Fabrizio Sebastiani, “Determining
term subjectivity and term orientation for opinipon
mining”, Proceedings of 11th Conference of the Eu-
ropean Chapter of the Association for Computational
Linguistics, Trento, Italy, 2006.
[29] Youngho Kim and Sung-Hyon Myaeng, “Opinion
Analysis based on Lexical Clues and their Expan-
sion”, Proceedings of NTCIR-6 Workshop Meeting,
pp. 308-315, Tokyo, Japan, 2007.
[30] Bo Pang and Lillian Lee, “A sentimental education:
Sentiment analysis using subjectivity summarization
based on minimum cuts”, Proceedings of 42nd Meet-
ing of the Association for Computational Linguistics,
pp. 271-278, Barcelona, Spain, 2004.
[31] Hugo, “MontyLingua: An end-to-end natural lan-
guage processor with common sense”,2003.
[32] Lun-Wei Ku, Yu-Ting Liang and Hsin-Hsi Chen,
“Opinion extraction, summarization and tracking in
news and blog corpora”, Proceedings of AAAI-2006
Spring Symposium on Computational Approaches to
Analyzing Weblogs, California, US, 2006.
[33] Sara Owsley, Sanjay Sood and Kristian J. Hammond,
“Domain specific affective classification of docu-
ment”, Proceedings of AAAI-2006 Spring Symposium
on Computational Approaches to Analyzing Web-
76
logs, California, US, 2006.
[34] Vasileios Hatzivassiloglou and Kathleen R.
McKeown, “Predicting the semantic orientation of
adjectives”, Proceedings of 35th Annual Meeting of
the Association for Computational Linguistics, pp.
174-181, Madrid, Spain, 1997.
[35] Wang and Araki, “Modifying SO-PMI for Japanese-
Weblog Opinion Mining by Using a Balancing Factor
and Detecting Neutral Expressions”, Proceedings of
NAACL HLT 2007, pp. 189-192, Rochester, New
York, US, 2007
[36] Alistair, Kennedy, Diana, Inkpen: Sentiment Classifi-
cation of Movie and Product Reviews Using Contex-
tual Valence Shifters. In: Proceedings of FINEXIN
2005, Workshop on the Analysis of Informal and
Formal Information Exchange during Negotiations.
Canada, 2005.
Anil Kumar.K.M is a faculty member with Department of Computer
Science and Engineering, Sri Jayachamarajendra College of Engi-
neering, Mysore. He is also a Research Scholar, Department of
Studies in Computer Science, University of Mysore. He received his
B.E. degree in 1999 from University of Mysore and M.Tech degree in
2006 from Visvesvaraya Technological University, Karnataka. He is
working towards doctoral work under the supervision of Dr. Suresha.
Anil Kumar.P is a Post Graduate scholar in the Department of
Computer Science and Engineering, Sri Jayachamarajendra College
of Engineering, Mysore, Karnataka, India. He received his Bachelor’s
degree in Computer Science and Engineering in 2002 from Visves-
varaya Technological University, Karnataka, India.
Dr. Suresha is presently working as Reader, Department of Studies
in Computer Science, University of Mysore. He received his B.Sc.
degree in 1987 from University of Mysore, M.Sc degree in 1989 from
University of Mysore and M. Phil degree in 1991 from DAVV, Indore.
He received M.Tech degree from Indian Institute of Technology, Kha-
ragpur in 1996 and Ph. D. from prestigious Indian Institute of
Science, Bangalore in 2007. He was awarded second prize in IRISS-
2002 competition, which is an all India level research student compe-
tition called “Inter Research Institute Student Seminar”. He has a
teaching experience of 20 years at post graduate level, has 12 publi-
cations to his credit and currently supervising four research scholars
towards their doctoral work.