Product Aspect Ranking using Sentiment Analysis: A Survey

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 03 Issue: 01 | Jan-2015 www.irjet.net p-ISSN: 2395-0072
© 2016, IRJET ISO 9001:2008 Certified Journal Page 124
Product Aspect Ranking using Sentiment Analysis: A Survey
Chetan Mate
1Department of Computer Engineering, MIT College of Engineering, University of Pune, India
Abstract: In today’s world, internet is the main origin
of information. There are many ecommerce websites
available where people discuss on different issues of
product. All ecommerce website provide facility to the
consumer to give opinion about their product and
services. Consumer reviews contain rich and valuable
knowledge for both firms and users. The problem with
this information is that these reviews are mostly
unorganized therefore creating difficulty for
information transfer knowledge accession. We propose
a product aspect ranking framework, which
automatically determine the important aspects of
products from online consumer reviews, aiming at
improving the usability of the countless reviews. The
important aspects are identified by two observations,
a) The important aspects of a product are usually
given by a large number of consumers; b) And
consumers’ opinions on the important aspects
influence their overall opinions on the product.
However Identifying important product aspects will
increase the usability of numerous reviews and is
useful to both consumers and firms. It is impractical for
people to manually identify the important aspects of
products from numerous reviews. Consumers can
conveniently make purchasing decision by paying more
attention to the important aspects, while enterprise
can focus on improving the quality of these aspects and
thus enhance product ranking effectively.
Keywords: e-commerce, aspect ranking,
product aspect, sentiment analysis
1. INTRODUCTION
Recent years have witnessed that in today’s
modern day life internet and web application are
playing very important role. Web application
provides simple way for people to do several task of
daily life like browsing, doing online transaction and
purchasing product. Due to the evolution and
advancement of information technology enabled
services many e-commerce websites are available,
hence result into more and more number of products
are sold on the internet. Thousands of products from
various dealers have been offered online. For
example, Bing Shopping has indexed more than five
million products. Amazon.com registers a total of
more than 36 million products. Shopper.com records
more than five million products from over 3,000
dealers [1]. Thus this affects more and more number
of people for purchasing products online. Over the
recent year there is quick growth and emergence of
e-commerce technology, inspire customer to buy
product online and express opinions on all kind of
objects such as product and services. In order to feel
customer more pleasant and more secure about
online shopping, it has become a conventional
practice for online dealers to enable their customers
to write reviews on product aspects that they have
purchased. Thus people not only buy product online
shopping but also they give review about product
aspect. As a result, the number of reviews that a
product receives grows rapidly. Here, an aspect, also
called feature in literatures, refers to a component or
an attribute of a certain product. A sample review
“The battery backup of Sony Xperia C mobile phone is
amazing." reveals positive opinion on the aspect
“battery backup" of product Sony Xperia C mobile
phone.
These customer reviews are very important
for social influence as well as an economic effect.

These reviews are useful to users for making
intelligent decisions about product purchasing and
also helpful to merchants for knowing their product’s
positive and negative attribute. In particular,
customer express sentiments (opinion) on various
aspects of products .A sentiment can be defined as
opinion expressed by consumer. Sentiments
represent any viewpoints of consumer such as like or
desirable (positive), dislike or undesirable (negative)
and may be neutral viewpoint. Sentiment
classification can be done at three levels of extraction
which are Document level, Sentence level and Aspect
level [9]. Such numerous consumer reviews contain
rich and precious knowledge and have become an
important source for both consumers and firms [1].
Consumers commonly find quality information from
online reviews prior to purchasing a product, while
many firms use online reviews as important
feedbacks in their product evolution, marketing, and
consumer relationship management. We argue that
some aspects are more important than the others,
and have greater impact on the eventual consumers’
decision making as well as firms’ product
development strategies. Hence, determining
important product aspects will improve the usability
of countless reviews and is useful to both consumers
and firms. Consumers can conveniently make
purchasing decision by paying more attentions to the
important aspects, while firms can focus on
improving the quality of these aspects and thus
enhance product reputation effectively. We propose a
product aspect ranking framework to automatically
identify the important aspects of products from
online consumer reviews.
In our proposed work, based on consumer
reviews of product first we recognize the important
features of product. Then we classify the sentiment
on that aspect. And then we develop the aspect
ranking algorithm for providing the rating to
particular product.
2. LITERATURE SURVEY
In this section, we review existing works
related to the Proposed framework. We start with the
works on the aspect identification from the consumer
review. Existing methods for aspect identification
based on supervised learning approaches and the
lexicon-based approaches. The lexicon-based
methods utilize a sentiment lexicon consisting of a
list of sentiment words, phrases and idioms, to
determine the sentiment orientation on each aspect
[14]. The supervised learning methods learn a
sentiment classifier based on training dataset. The
classifier is then used to predict the sentiment on
each aspect. Many learning-based classification
models are applicable, for example, Support Vector
Machine (SVM), Naive Bayes, and Maximum Entropy
(ME) model etc. [7]. Supervised learning is based on
the training data and cannot perform well without
sufficient training samples. However, labeling
training data is tedious and time-consuming. In this
work, the Pros and Cons reviews have explicitly
categorized positive and negative opinions on the
aspects. Most existing supervised methods are based
on the sequential learning technique. For example,
Wong and Lam [7] learned aspect extractors using
Hidden Marköv Models and Conditional Random
Fields, respectively. Jin and Ho [11] learned a
lexicalized HMM model to extract aspects and
opinion expressions, while Li et al. [9] integrated two
CRF variations, i.e., Skip-CRF and Tree-CRF. On the
other side, unsupervised methods have emerged
recently. The most notable unsupervised approach
was proposed by Hu and Liu [12]. They assumed that
product aspects are nouns and noun phrases. The
approach first extracts nouns and noun phrases as
candidate aspects. The occurrence frequencies of the
nouns and noun phrases are counted, and only the
frequent ones are kept as aspects. The next task is
aspect sentiment classification, which determines the
orientation of sentiment expressed on each aspect.
Two major approaches for aspect sentiment
classification include lexicon-based and supervised
learning approaches. The lexicon-based methods are
typically unsupervised. They rely on a sentiment
lexicon containing a list of positive and negative
sentiment words. To generate a high-quality lexicon,
the bootstrapping strategy is usually employed. For
example, Hu and Liu [12] started with a set of
adjective seed words for each opinion class (i.e.,
positive or negative). They utilized
synonym/antonym relations defined in WordNet to
bootstrap the seed word set, and finally obtained a
sentiment lexicon.

Document-level sentiment classification aims
to classify an opinion document as expressing a
positive or negative opinion. Existing works use
unsupervised, supervised or semi-supervised
learning techniques to build document level
sentiment classifiers. Unsupervised method usually
relies on a sentiment lexicon containing a collection
of positive and negative sentiment words. It
determines the overall opinion of a review document
based on the number of positive and negative terms
in the review. Supervised method applies existing
supervised learning models, such as SVM and
Maximum entropy (ME) etc. [5], while semi-
supervised approach exploits abundant unlabeled
reviews together with labeled reviews to improve
classification performance. The two widely used
methods are the sentence ranking and graph-based
methods [10]. In these works, a scoring function was
first defined to compute the informativeness of each
sentence. Sentence ranking method [9] ranked the
sentences according to their informativeness scores
and then selected the top ranked sentences to form a
summary. Graph-based method [7] represented the
sentences in a graph, where each node corresponds
to a sentence and each edge characterizes the
relation between two sentences. A random walk was
then performed over the graph to discover the most
informative sentences, which were in turn used to
compose a summary.
3. PROPOSED WORK
Following four steps are used in proposed work.
1)Reviews extraction and Preprocessing.
2)Aspect Identification of the product
3)Classify the positive and negative reviews of
product by sentiment classifier.
4)The probabilistic ranking algorithm used for
product ranking.
3.1 Reviews Extraction and Preprocessing
First step of product aspect ranking
framework is data preprocessing which is a very
important step. Compared to formal text document
the reviews are generally less formal and written in
particular manner. If the sentiment analysis applied
on unstructured review generally achieve very poor
performance in most case. Therefore the
preprocessing techniques on reviews are important
for obtaining acceptable result on sentiment analysis.
Following are the methods of preprocessing.
1)Stemming: In stemming we will remove the postfix
from each word such as ing,tion etc. Eg. Buying will
become buy after stemming.
2)In tokenization we will tokenize each sentence by
space. Means we will remove the spaces. Stop word
removation like a, an, the etc.
Figure 3.1: Architecture Diagram
3.2 Aspect Identification of the product
Aspect identification is second important
step in which we identify aspect from countless
consumer reviews. The reviews are available on
different ecommerce websites. But the problem
with these reviews is that they are composed in
different formats on various websites. Consumer
review consists of positive and negative reviews. On
Consumer Review
Preprocessing
Aspect
Identification
Sentiment
Classification
Aspect Ranking
Overall Rating

some website there is a overall grading on the
product, some website the reviews are in paragraph
in formal text. e.g. CNet.com, Viewpoints.com,
Reevoo.com are the various websites for reviews of
different product and has different formats. From
these reviews the aspect are determine as a frequent
Noun term and noun phrases. We can get precise
aspect by extracting persistent noun from the
positive and negative reviews[3]. Hu and Liu
proposed most remarkable approach for aspect
identification. In this approach it first identifies the
noun and noun phrases in the document. The
occurrence frequency of noun and noun phrase are
counted then only the consistent noun terms are kept
as the aspect [4].Phrase dependency parser used to
extract the noun phrase by Wu et al. Some aspect
may contain synonym term such as “headphone” and
“earphone.”In such situation we perform synonym
clustering to obtain unique aspect. These synonym
terms are obtained from synonym dictionary
Websites.
3.3. Sentiment Classifier
Sentiment analysis or Opinion mining is a
type of natural language processing used for tracking
the mood o r polarity of people about product aspect.
Sentiment classification used to classify the given text
to one or more predefined sentiment categories. Such
as Positive, Negative, Neutral. There is various
classification techniques are available..There are two
types of learning supervised learning and another is
unsupervised learning. The lexicon-based approaches
are mostly unsupervised. The lexicon based methods
make use of a sentiment lexicon consist of list of
sentiment words, phrases and idoms, to obtain
sentiment polarity on each aspect[9].the
performance of supervised learning dependent on
training data. It not able performs well without
sufficient data. Supervised learning method train a
sentiment classifier based on training dataset. The
classifier is used to determine the sentiment on each
aspect. There are many learning based classification
models are available. Support Vector Machine (SVM),
Naive Bayes, and Maximum Entropy (ME)model
these are the learning based classification model[10].
Among all these learning based classification
methods Support Vector Machine (SVM) having
significant accuracy.
3.4 Aspect Ranking Algorithm
In proposed framework important aspects
are commented frequently in consumer review and
the consumer’s opinions on the important aspects are
greatly affects their overall opinions on the product.
The overall opinion in a review is an aggregation of
the opinions given to specific aspects in the review,
and various aspects have different contributions in
the aggregation. That is, the sentiment on
(un)important aspects have strong (weak) impacts
on the generation of overall opinion. In proposed
framework we use AFINN dictionary approach.
AFINN is a list of English words rated valence with an
integer between minus five (negative) and plus five
(positive).
4. APPLICATIONS
Aspect ranking framework is useful to a huge
number of real world applications. Businesses and
organizations are interested in opinions for use them
for new product perception, brand perception,
reputation management, product and service
benchmarking, market intelligence. Consumers are
interested in other’s opinions when purchasing a
product, using a service, tracking political topics etc.
CONCLUSION
In this paper, we have surveyed the reference
papers related to Aspect identification, Sentiment
classification. We have planned to identify the
important aspects of a product from online consumer
reviews. Our assumption is that the important
aspects of a product should be the aspects that are
frequently commented by consumers and consumers’
opinions on the important aspects greatly pressure
their overall opinions on the product. Based on this
assumption, we will try to develop an aspect ranking
algorithm which will identify the important aspects
by concurrently considering the aspect frequency
and the pressure of consumers’ opinions given to
each aspect on their overall opinions.

REFERENCES
[1] Zheng-Jun Zha, Jianxing Yu, Jinhui Tang, Meng
Wang, Tat-Seng Chua,” Product Aspect Ranking and
Its Applications”.IEEE2014.
[2] B. Liu, “Sentiment analysis and subjectivity,” in
Handbook of Natural Language Processing, New York,
NY, USA: Marcel Dekker, Inc., 2009.
[3] B. Liu, M. Hu, and J. Cheng, “Opinion observer:
Analyzing and comparing opinions on the web,” in
Proc. 14th Int. Conf. WWW, Chiba, Japan, 2005, pp.
342–351.
[4] M. Hu and B. Liu, “Mining and summarizing
customer reviews,” in Proc. SIGKDD, Seattle, WA, USA,
2004, pp. 168–177.
[5] Finn, A., Kushmerick, N., and Smyth, B. 2002.
Genre Classification and Domain Transfer for
Information
Filtering. In Proc. of European Colloquium on
Information Retrieval Research, pages 353-362.
[6] X. Ding, B. Liu, and P. S. Yu, “A holistic lexicon-
based approach to opinion mining,” in Proc. WSDM,
New York, NY, USA, 2008, pp. 231–240.
[7]T. L. Wong and W. Lam, “Hot item mining and
summarization from multiple auction web sites,” in
Proc. 5th IEEE ICDM, Washington, DC, USA, 2005, pp.
797–800.
[8] O. Etzioni et al., “Unsupervised named-entity
extraction from the web: An experimental study,” J.
Artif. Intell., vol. 165, no. 1, pp.91–134. Jun. 2005.
[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] W. Jin and H. H. Ho, “A novel lexicalized HMM-
based learning framework for web opinion mining,”
in Proc. 26th Annu. ICML,Montreal, QC, Canada, 2009,
pp. 465–472.
[12] M. Hu and B. Liu, “Mining and summarizing
customer reviews,” in Proc. SIGKDD, Seattle, WA, USA,
2004, pp. 168–177.
[13] K. Jarvelin and J. Kekalainen, “Cumulated gain-
based evaluation of IR techniques,” ACM Trans.
Inform. Syst., vol. 20, no. 4,pp. 422–446, Oct. 2002.
[14] B. Ohana and B. Tierney, “Sentiment
classification of reviews using SentiWordNet,” in
Proc. IT&T Conf., Dublin, Ireland, 2009.
[15] K. Lerman, S. Blair-Goldensohn, and R.
McDonald, “Sentiment summarization: Evaluating
and learning user preferences,” in Proc. 12th Conf.
EACL, Athens, Greece, 2009, pp. 514–522.

Product Aspect Ranking using Sentiment Analysis: A Survey

More Related Content

What's hot (14)

Similar to Product Aspect Ranking using Sentiment Analysis: A Survey (20)

More from IRJET Journal (20)

Recently uploaded (20)

Product Aspect Ranking using Sentiment Analysis: A Survey