Product aspect ranking using domain dependent and domain independent review

IJRET: International Journal of Research in Engineering and Technology eISSN: 2319-1163 | pISSN: 2321-7308
_______________________________________________________________________________________
Volume: 03 Issue: 11 | Nov-2014, Available @ http://www.ijret.org 160
PRODUCT ASPECT RANKING USING DOMAIN DEPENDENT AND
DOMAIN INDEPENDENT REVIEW
Priti Sole1
, M.C.Kshirsagar2
1
VACOE, Ahmednagar, Pune University
2
VACOE, Ahmednagar, Pune University
Abstract
In today’s world, internet is the main source of information. There are many blogs and forum sites available where people
discuss on different issues and also almost all ecommerce website provide facility to the users to express opinion about their
product and services which is important information available on the internet .The problem with this information is that this
reviews are mostly not organized therefore creating difficulty for knowledge acquisition. There are many solution exist to resolve
this problem but the available existing methods depends on extracting product aspect only considering single domain relevant
review corpus. To address this problem, a method is explored to identify product aspect from online review is by taking into
account the difference in aspect statistical characteristic across different corpus. This paper shows need of automatically
identifying important product aspects from available online customer review and an approach of aspect ranking. This paper
also shows the related work on this domain. Our methodology confirmed product aspect which are less nonspecific in domain
independent corpus and more domain specific. Then customer opinion expressed on these aspects is determined using sentiment
classifier and finally ranking of product aspect is calculated using it’s ranking relevance score of each aspect .
Keywords— Product aspect, aspect ranking, sentiment classification, customer review, opinion mining, aspect
identification, product ranking.
--------------------------------------------------------------------***------------------------------------------------------------------
1. INTRODUCTION
In today’s modern day life internet and web application are
playing very important role. Web application provides
easiest way for people to do several activities of daily life
like browsing, doing online transaction and purchasing
product. Due to the development and advancement of
information technology enabled services a lot of e-
commerce websites are available, hence result into more and
more number of products are sold on the web. Thus this
influence more and more number of people for buying
products online. Over the last year there is rapid growth and
emergence of e-commerce technology, motivate customer to
buy product online and express opinions on all kind of
objects such as product and services. In order to feel
customer more comfortable and more secure about online
shopping, it has become a common practice for online
merchants to enable their customers to write reviews on
products that they have purchased. Thus people not only buy
product online but also they give feedback about product. As
a result, the number of reviews that a product receives grows
rapidly For example of customer review is: “The picture
quality of Nikon D5000 Digital camera is good”.
These customer review has social impact as well as an
economic impact. The customer reviews in shopping web
sites are very much helpful for product advertisement in
which satisfied customers post their comment to know other
user what they like about the product and how much. These
reviews are useful to other users for making informed
decisions about product purchasing and also helpful to
merchants for knowing their product’s positive and negative
attribute.
Customer write review on various aspects of product. In this
paper the word aspect is used to represent any particular part
,attribute or feature of product. In particular, customer
express sentiments on various aspects of products A
sentiment can be defined as opinion expressed by reviewer.
In above example reviewer gives positive opinion on
specific aspect picture quality of product digital camera
.Sentiments represent any viewpoints of customer such as
like or desirable (positive),dislike or undesirable(negative)
and may be neutral viewpoint. For instance, sentiment
classification looks for emotional expression such as good,
bad excellent ,poor etc. Sentiment classification can be done
at three levels of extraction which are Document level,
Sentence level, Aspect level [9]. This mechanism mainly
focus on splitting review document at above specified levels
to determine expressed opinion whether it is positive
,negative or neutral. The task of summaries is clearly
different from traditional text summarization [1] because it
does not summarize the reviews by selecting or rewriting a
subset of original sentences from the reviews.
The fig.1 shows example of customer review on cell phone
LG G3. This review shows feedback on multiple aspects of
the cell phone LG such as display, camera and processor. It
means that product may have number of aspects. Figure
shows that the reviewer gives overall rating only for cell
phone LG without giving any individual rating on each
aspect. Hence user would not be able to draw any conclusion

_______________________________________________________________________________________
about the reviewer’s opinion on each aspect. Although
different reviewer may give a cell phone LG the same
overall rating, its reason of same overall rating may vary
according to different reviewer. For example, one reviewer
may like display quality of cell phone ,but other user may
like the quality of processor because may be heshe is
interested in processing power of cell phone. And also some
aspects has more importance as compared to other aspects.
For instance, for camera product aspect such as “lenses” and
“picture quality ” are concerned by most customers and are
more important than the other aspects such as “a/v cable ”
and “wrist strap”. Therefore it is necessary to tell other users
different reviewers individual rating on each important
aspect because customer opinion on an important aspects of
product influence his or her overall opinion/rating of
product. Yet customer provide individual rating on each
aspect it is difficult for user to manually determine rating on
important aspect because of rapidly increasing large volume
of online review.
Fig.1. Example of customer review on Cell phone LG G3
Straight we can say that frequently commented aspects are
more important but this is not the true. Frequency based
solution is not able to determine truly important aspects
because sometimes customer opinion on the frequently
commented aspect may not affect his/her overall opinion.
For example , reviewer frequently criticize on “bad exterior
” of HTC but still give high overall rating .So, we can not
say that highly commented aspect in review is truly
important .We need another kind of solution to find out
important aspect. This is the motivation of this paper. This
motivate us to present a new approach to automatically
determine the important product aspects and rank aspect
according to it’s importance score so that user can easily
understand what are important aspect of product.
2. RELATED WORK
In this section, we present some of the existing research and
related work in the aspect based opinion mining :
Existing product aspect extraction approaches can be
divided into two categories, called, supervised and
unsupervised. Supervised method requires set of labeled
reviews as training example. A supervised learning method
is then applied to make an extraction model and then it is
capable of identifying product aspect. Various technique
such as Hidden Markov Model and Conditional Random
Field[3,4],Maximum entropy[2] are used for this task.
However, supervised methods are time consuming because it
requires time for preparing training examples.
M. Hu and B. Liu[5]2004 proposed unsupervised method.
In this paper they showed that their proposed summarization
task is different from traditional method because they
determine only those aspects of product on which opinion is
expressed. They consider that product aspect are generally
noun or noun phrases. NLProcessor linguistic parser is used
to do part of speech tagging to determine syntactic structure
of sentence that determines whether a word is
noun,verb,adjective etc..Thus they identified noun or noun
phrase which identified as aspect and those aspects which
are frequently commented by user finally determined.
Y. Wu, Q. Zhang, X. Huang, and L. Wu[7] 2009 Explored
phrase dependency parser to extract noun phrases from
review as candidate aspect They observed that a lot of
product aspects are phrases.Firltly they identify dependency
grammar structure of sentence determining relation between
head and its dependent. Then phrase dependency tree is
constructed from the result obtained in first stage. Finally
candidate product aspect are identified opinion expressed on
this aspect is also determined.
Sentiment classification task focus on determining semantic
orientation on each aspect ie.positive,negative or neutral.
Sentiment analysis conducted at one of the three level
:Document level, Sentence level and aspect level. Sentiment
classification has mainly two approaches that are lexicon
based and supervised learning . Lexicon based methods are
unsupervised and they depend on sentiment lexicon
containing desirable and undesirable words .In contrast
supervised method determine the opinion on aspects by
using sentiment classifier.
B.Pang,L.Lee,and S.Vaithyanathan [4] 2002 proposed
three machine learning methods, naïve Bayes,maximum
entropy and support vector machine to classify whole movie
reviews into positive or negative opinion. They conclude
that standard machine learning methods produced good
result as compared to human-generated baseline.They also
showed that naïve bayes gives worst result whereas support
vector machine gives best result. In most of the comparative
studies it is found that support vector machine outperforms
other machine learning methods in sentiment classification.
P.D.Turney[8]2002 proposed unsupervised method to
classify review documents as recommended(positive) and
not recommended (negative) in. In this paper Pointwise
Mutual Information (PMI) and Information Retrieval (IR)
algorithm is used to measure semantic orientation of word.
A semantic numerical score of word is calculated by
considering mutual information between the given word and
already defined positive word and subtracting mutual
information between given word and predefined negative
word. Depending on average semantic numerical score the

_______________________________________________________________________________________
review document get final remark as either positive or
negative. But the disadvantage of this method lies in time
required to execute queries and accuracy in another
application.
F.Li[6] 2010 in this paper, they focus on object feature
based review summarization. In this paper they proposed a
new methlodology based on Conditional Random Fields.
T. Wilson, J. Wiebe, and P. Hoffmann[13]2005 presented
an approach to predicting contextual sentiments at the phrase
level by applying machine learning techniques on a variety
of feature factors. First they determine whether opinion
expression is neutral or not. Then they distinguish sentiment
polarity into positive, negative or neutral opinion at phrase
level.
Yongyong Zhail, Yanxiang Chenl, Xuegang [14] (2010)
attempted to create a novel framework for sentiment
classifier learning from unlabeled documents. The process
begins with a collection of un-annotated text and a sentiment
lexicon. An initial classifier is trained by incorporating prior
information from the sentiment lexicon which consists of a
list of words marked with their respective polarity. The
labeled features use them directly to constrain model‟s
predictions on unlabeled instances using generalized
expectation criteria. The initially-trained classifier using
generalized expectation is then applied on the un-annotated
text and the documents labeled with high confidence are fed
into the self-learned features extractor to acquire domain-
dependent features automatically. Such self-learned features
are subsequently used to train another classifier which is
then applied on the test set to obtain the final results.
H. Wang, Y. Lu, and C. X. Zhai.[12] 2010 developed a
latent rating regression analysis model.Advanatge of this
method is that we are able to find out latent rating of each
aspect from given text review and overall rating of product.
First they find out major aspects by using bootstrapping
based algorithm. They assume that overall rating is weighted
aggregation of underlying rating on each aspect and it’s
weight. Weight is nothing but the importance placed by
customer on each aspect. Then latent regression analysis
model is used to find out individual reviewer’s underlying
ranking on each major aspect and the relative important
weight on different aspects. Limitation of this model is that
they concentrate on reviewer rating behavior analysis rather
than on aspect ranking.
3. WORKING OF SCHEME
Customer review are in the form of unstructured text format.
Customer review contains different kind of opinion on
different aspects of products. Opinion mining is used to find
out aspects from given review corpus and deriving opinion
on it. In this paper we present a method which automatically
identify important aspects. There are many techniques exist
for opinion based mining. But each technique has it’s own
consequences as shown in related work. Generally existing
techniques depends on mining patterns only from single
review corpus. The framework contains three main
mechanism ie. Product aspect identification, aspect
sentiment classification and product aspect ranking.
For the task of aspect extraction unsupervised natural
language processing method is used. This method defines
domain independent specific template. We will not only
focus on domain dependent corpus but also domain
independent corpus. A set of candidate aspects are extracted.
Then for each aspect we find out it’s DDR(Domain
Dependent Relevance)and DIR(Domain Independent
Relevance) score is calculated.DDR represent statistical
relation of the candidate aspect to given review corpus and
on the other hand DIR represent statistical relation of the
candidate aspect to the domain irrelevant corpus. Aspects
that has less DIR score means less non specific in given
domain independent corpus and more DDR score means
more domain specific are confirmed as candidate aspect.
Then next task is to determine semantic expressed on
extracted candidate aspect. Sentiment classifier is used to
perform this task which classifies semantic orientation on
each aspect.
Then probabilistic ranking algorithm is used to find out the
ranking score of various aspect of product from numerous
review. The algorithm consider aspect frequency and take
into account relation between the overall opinion and the
opinions on specific aspects. The opinions on important
aspects have strong impacts on the generation of overall
opinion and on the other hand opinions on unimportant
aspects have weak impacts on the generation of overall
opinion. By taking into consideration above fact ranking
score of each aspect is calculated and then product aspect are
finally ranked according to it.
4. CONCLUSION
In this paper ,we have shown the need of determining
important product aspect from available online numerous
reviews and an approach of product aspect ranking. Also we
have shown related work and existing approaches to this
domain. In this paper, a method for identifying product
aspects from customer reviews has been presented. First of
all, the candidate product aspects are identified which are
specific to the given review domain and yet not
general(domain independent). Customer opinion on these
aspects are determined using sentiment classifier. Finally,
the identified product aspects are ranked according to their
relevance score.
ACKNOWLEDGMENTS
This paper work is completed successfully only because
support from each and every one including teachers,
colleague, parents and friends. Especially, I am very
thankful to those who provide me guidance and make this
work reachable. This paper work is supported by my senior,
my teachers and some experienced personalities.My
acknowledgment of gratitude toward the everyone who
make this work reachable.

_______________________________________________________________________________________
REFERENCES
[1] Aurangzeb Khan,Baharum Baharudin and
Khairullah Khan, “Sentiment Classification Using
Sentence-level Lexical Based Semantic Orientation
of Online Reviews.” Trends in Applied Sciences
Research, 6: 1141-1157, 2011.
[2] blogs," in International Conference on Weblogs and
Social Media (ICWSM), pp. 219-222.
[3] M. Hu and B. Liu, "Mining and summarizing
customer reviews," presented at the Proceedings of
the tenth ACM
[4] B.Pang,L.Lee,and S.Vaithyanathan,” Thumbs up?
Sentiment classification using machine learning
techniques,”in Proc EMNLP, Philadelphia, PA,
USA, pp. 79–86,2002.
[5] M. Hu and B. Liu, “Mining and summarizing
customer reviews,” in Proc. SIGKDD, Seattle, WA,
USA, pp. 168–177,2004.
[6] F. Li et al., “Structure-aware review mining and
summarization,” in Proc. 23rd Int. Conf. COLING,
Beijing, China, pp. 653–661,2010.
[7] Y. Wu, Q. Zhang, X. Huang, and L. Wu, “Phrase
dependency parsing for opinion mining,” in Proc.
ACL, Singapore, pp. 1533–1541,2009.
[8] P.D. Turney, “Thumbs Up or Thumbs Down?:
Semantic Orienta- tion Applied to Unsupervised
Classification of Reviews,” Proc. 40th Ann. Meeting
on Assoc. for Computational Linguistics, pp. 417-
424, 2002.
[9] Bing Liu, “Sentiment Analysis and Opinion Mining”
pp.7-140,2012.
[10] V. Gupta and G. S. Lehal, “A survey of text
summarization extractive techniques,” J. Emerg.
Technol. Web Intell., vol. 2, no. 3, pp. 258–268,
2010.
[11] D.M. Blei, A.Y. Ng, and M.I. Jordan, “Latent
Dirichlet Allocation,” J. Machine Learning
Research, vol. 3, pp. 993-1022, Mar. 2003.
[12] H. Wang, Y. Lu, and C. X. Zhai, “Latent aspect
rating analysis on review text data: A rating
regression approach,” in Proc. 16th ACM SIGKDD,
San Diego, CA, USA, pp. 168–176,2010
[13] T. Wilson, J. Wiebe, and P. Hoffmann,
“Recognizing Contextual Polarity in Phrase-Level
Sentiment Analysis,” Proc. Conf. Human Language
Technology and Empirical Methods in Natural
Language Processing, pp. 347-354, 2005.
[14] Yongyong Zhail, Yanxiang Chenl, Xuegang Hu,
“Extracting Opinion Features in Sentiment Patterns”
, International Conference on Information,
Networking and Automation (ICINA),2010 .

Product aspect ranking using domain dependent and domain independent review

More Related Content

What's hot (17)

Viewers also liked (20)

Similar to Product aspect ranking using domain dependent and domain independent review (20)

More from eSAT Publishing House (20)

Recently uploaded (20)

Product aspect ranking using domain dependent and domain independent review