www.ijecs.

in
International Journal Of Engineering And Computer Science ISSN: 2319-7242
Volume 6 Issue 1 Jan. 2017, Page No. 19965-19968
Index Copernicus Value (2015): 58.10, DOI: 10.18535/ijecs/v6i1.20

Sentiment Analysis of Reviews for E-Shopping Websites

Dr. U Ravi Babu1
1
Professor, Dept of Computer Science and Engineering,
Narsimha Reddy Engineering College,
Secunderabad, TS, INDIA
[email protected]
Abstract: The sentiment analysis is one of the popular research area in the field of text mining. Internet has become very popular resource
for information gathering. People can share their opinion related to any product, services, events etc over internet. Websites like Amazon,
Snapdeal, Homeshop18 etc are popular sites where millions of users exchange their opinions and making it a valuable platform for
tracking and analyzing opinion and sentiments. “What other people thing” is being an important piece of information whenever we want to
take any decision. Sentiment analysis is the best solution. This gives important information for decision making in various domains.
Various sentiment detection methods are available which affect the quality of result. In this paper we are finding the sentiments of people
related to the services of E-shopping websites. The main goal is to compare the services of different E-shopping websites and analyzing
which one is the best. For this we use five large dataset of five different E-shopping website which contains reviews related to the services.
“Sentiwordnet dictionary” is used for finding scores of each word. Then sentiments are classified as negative, positive and neutral. It has
been observed that the pre-processing of the data is greatly affecting the quality of detected sentiments. Finally analysis takes place based on
classification.
Keywords: Sentiment analysis, opinion mining; E-shopping websites; classification.
 Now finding the sentiments for the five large E-
1. Introduction shopping website dataset. It uses “sentiwordnet
dictionary” for finding score of each words.
Sentiment analysis is one of the current research topics in  Then sentiments are classified as positive, negative and
the field of text mining. Opinions and sentiments mining from neutral.
natural language are very difficult task. Sentiments are  We analyze how preprocessing techniques and type of
extracted from comments, reviews, feedbacks etc. “What other input data can affect the quality of topic detection
people think” has always been an important piece of method.
information while taking any decision. Now days, before Finally analysis takes place based on classification. The
planning to go for movie, everyone what to know its reviews. analysis of the services according to positive and negative
Sentiment analysis is the best tool for finding whether the reviews can be shown in the graphical format.
review is positive or negative. It helps people to find good
quality product. It also helps companies by providing The rest of the paper is prearranged as follows. Section II
customers feeling related to their product. It also helps to presents some previous research work and challenges of
analyze public sentiments related to political issues or political sentiment analysis. Section III presents the description of
candidates. The main focus of the system is to analyze dataset, some pre-processing steps and sentiments analysis
sentiments for E-shopping company services. The reviews are techniques which we use in this work. In Section IV, the
classified according to positive, negative and neutral score. Experimental results are shown and in section V conclusion is
These results can guide us to select particular site for e- given.
shopping, based on maximum number of positive reviews.
So for detecting and analyzing sentiments, we need the 2. Related Work
streams of data generated from online sources. The first step is
to collect the reviews related to services provided by company. Shulong Tan et al., had proposed a two LDA based model
We are using five dataset collected from online sources. We (FB-LDA and RCB-LDA) for analyzing public sentiments
select five popular e-shopping website for our work: Amazon, variations and finding the possible reasons causing this
Flipkart, Home Shop 18, Jabong, and Snapdeal. All these sites variation. Their work mainly focused on tracking sentiments
are very popular in present from where most of the people like and interpreting sentiment variation. Proposed model also used
to purchase. for finding topic differences between two sets of document. [1]
LIU Lizhen et al., had proposed a feature-based vector model
In short, our work can be summarized as follows:
and a novel weighting algorithm for sentiment analysis of
 Firstly we collect the five large E-shopping websites
Chinese product reviews. They classify reviews into positive,
dataset which contains review related to the services
negative and neutral comments. They used supervised
of particular websites.
sentiment classification and a novel feature weighting
 Then we apply some preprocessing techniques on algorithm [2].
datasets for removing unwanted things and arranging Jalaj S. Modha et al., discussed about exiting approaches
data in proper manner.
methods etc. for performing sentimental analysis on
 After that we use POS tagger for assigning tags to each
unstructured data available on web. Previously, Sentiment
word according to its role.
Analysis concentrated for subjective statements or on
subjectivity and it just overlooked objective statements which
Dr. U Ravi Babu, IJECS Volume 6 Issue 1 Jan., 2017 Page No.19965-19968 Page 19965
 DOI: 10.18535/ijecs/v6i1.20

carry sentiment(s). They proposed a new approach which topic detection methods are applied on raw comments then it
classifies and handles not only subjective but also objective frequently get very poor performance. Therefore for removing
statements for sentimental analysis. They evaluated their noise and unwanted things we use different preprocessing
experimental results by using information Retrieval matrices techniques. They are very important for obtaining satisfactory
such as precision, recall, f-measure and accuracy [3]. results. Different preprocessing techniques we used are as
Subhabrata Mukherjee et al., presented a novel approach follows:
which identified feature specific expressions of opinion in Tokenization: Tokenization is the process of extracting bags of
product reviews with different features and mixed emotions. [4] cleaner terms from raw comments by deleting stop words and
M. Thelwall et al., and Y. Tausczik et al., describe the punctuation, compressing redundant character repetitions and
SentiStrength tool which is based on LIWC sentiments lexicon. deleting IDs or name used in the text for messaging purposes.
These two tools are used to assign sentiment labels for each For removing stop word we maintain stop word dictionary
tweet. [5] [6]. which contain all stop words. We compared each word of
Pang et al., presented a detailed survey of the existing comments with this dictionary and the matched word gets
methods on sentiment analysis. Sentiment analysis, also known removed from comment.
as opinion mining which are widely applied to various
Slang words translation: User generated comments often
document types, such as movie or product reviews. Online
contains the slang words. Slang word translation means
public sentiment analysis is gradually more popular topic in
converting the slang words like lol, omg etc, into their standard
social network related research. There has been some research
work focusing on assessing the relations between online public form. We used the Internet Slang Word Dictionary for this and
sentiment and real-life events. [7] then add them to the comments.
Pimpalkar, et al., had developed a system that shows the Stemming: Stemming means a group of different words share
comments and feedbacks/reviews for products. They the same meaning. It is the process of reducing words which
determined the polarity of sentiments for the products‟ reviews share the same meaning. We used stem word dictionary for
of the person. Finally they showed the prediction about grouping all different words of same meaning.
product. This comparison leads to find the best product. The URL removal: Many users include URLs in their tweets. These
rule based and fuzzy logic approach was used to give the URLs make the sentiment analysis process more complex. So
output. [8] URLs are removed from the tweets
C. Sentiment Analysis
3. Sentiment analysis From E-Shopping Services  POS Tagging: “POS (Parts of speech) tagging” means
reviews a type to which a word is assigned in according to its
Next, we describe all the components of our related work. syntactic functions. In English language the main “parts
In Section III-A, we describe the dataset. Then in Section III-B of speech” are pronoun, noun, adjective, verb, adverb,
we explain the data preprocessing techniques. In Section III-C etc. “POS tagging” means assigning the labels (tags) to
we used different sentiment analysis techniques that take words in sentence according to its function in the
preprocessed data as input and classifying them as positive, sentence. In our work for assigning a label (tag) to each
negative and neutral comments. word, we used “Stanford POS (Parts of Speech) tagger”.
A tag is allocated to each word, like, NNS, NN, JJS, JJ,
A. Datasets
RB, VB etc.
We describe dataset used in this work. We prepared dataset  Sentiments classification: In this we calculate the
from online e-shopping websites. We collected reviews polarity of the sentiments and classify them as
related to services of these websites. We selected five positive, negative and neutral. We used
popular e-shopping website for our work: Amazon, “SentiWordNet dictionary” for calculating the
Flipkart, Home Shop 18, Jabong, and Snapdeal. All these score of each word. „SentiWordNet” is a lexical
sites are very popular in present from where most of the resource in which each word is associated to
people like to purchase. We collected near about 2000 positive, negative and objective score. So with the
reviews for each site. Then we arranged our dataset in help of these score we find the positive, negative
required format. Below table show the number of reviews and objective score of each word in review
collected for experimental work. comments. We use below simple logic for finding
whether the word is positive, negative or neutral.
Table 1: Details of the Dataset
Category Name of sites No of comments
If Pos_Score of word > Neg_Score of word;
Amazon 2000 We consider word is positive;
E-shopping website Else If Pos_Score of word < Neg_Score of word;
services related Flipkart 1900 We consider word is negative;
reviews Snapdeal 1800 Else If Pos_Score == Neg_Score;
Home shop 18 1900 Word is neutral;
Jabong 2000 End If;
Then we perform summation of all positive score words and
B. Data Preproccesing negative score words for knowing the status of whole comment
whether it is positive or negative. After performing summation
User generated massages are very noisy. They are less we calculate the final positive and final negative score with the
formal and generally used non English words and symbols. If help of below given logic.

Dr. U Ravi Babu, IJECS Volume 6 Issue 1 Jan., 2017 Page No.19965-19968 Page 19966
 DOI: 10.18535/ijecs/v6i1.20

Final_Pos_score = SPos / ( SPos + SNeg); Flipkart Flipkart is my all Flipkart is my Flipkart my

Final_Neg_score = SNeg / ( SPos + SNeg); time favorite when it all time favorite time favorite
Where SPos is a variable which contains summation of all comes to online when it come to come online
shopping. It has online shopping. shopping
positive score words and SNeg contains summation of all
made shopping one It has make made shopping
negative score words of my favorite shopping one of favorite
If ( Final_Pos_Score – Final_Neg_Score ) > 0.1 activities and I can my favorite activitie
Whole comment is positive; indulge into this for activitie and I indulge hours
ElseIf (Final_Neg_Score – Final_Pos_Score)> 0.1 hours and hours. can indulge into hour exchange
Whole comment is negative; The exchange policy this for hours policy offer
and offers are good. and hour. The good Flipkart
ElseIf Comment is neutral;
Flipkart is good for exchange policy good electronic
Finally obtain all positive, negative and neutral comments. the electronics and offer are device
The positive, negative and neutral feedbacks will be classified devices. good. Flipkart is
further for analysis. good for the
electronic
device.
4. Experimental Results
To detect the sentiments of people related to services of E- Table 3: Example Of Pos Tagging Ans Sentiwordnet Dictionary
shopping websites we generate five large dataset which
contains reviews for services. We collected these reviews from
online sources. The selected dataset contains the reviews of
most popular e-shopping websites: Amazon, Snapdeal, Home
Shop 18, Flipkart, and Jabong.
Tables II show the result of preprocessing in which results of
preprocessing methods on Amazon and Flipkart dataset are
given. In comment field reviews related to Amazon and
Flipkart are given. We stem all different words which share
same meaning and output is given in After_Stemming field.
After that we removed all unwanted things like symbols
punctuation, stop words, removing mentions and compressing
repeated words from the tweets and the output is given in
After_Stoping field. Table 4: Example of Sentiments Classification
In Table III the results of POS tagger and SentiWordNet
dictionary are shown. A tag is allocated to each word, like, NNS,
NN, JJS, JJ, RB, VB, etc using POS tagger. To calculate score of
an individual word SentiWordNet dictionary is used. Calculated
score for each word are shown. Then we add all positive score and
all negative score and find the total positive score and total
negative score of whole comment.

In Table IV the result of classification of positive, negative and

neutral comments are shown. From this we can analyze the
views of people related to services of companies. How many The figure 1 shows the graphical analysis of the proposed
are in favour and how many are in oppose related to services method.
provided by particular company.

Table 2: Example of Preprocessing

Name Comment After_Stemming After_Stopping
Amazon Low prices fast and Low prices fast Low price fast
reliable delivery and reliable reliable
AND vast selection. delivery AND delivery vast
Customers can now vast selection. selection
buy products any Customer can Customer buy
time. Realy Good now buy product time
product fair pricing product any Realy Good
overall amazon is time. Realy product fair
good website to buy Good product price overall
branded product on fair price overall amazon good
fair prices. amazon is good website buy
website to buy brand product
brand product fair price
on fair price.
Fig.1. Result of graphical analysis.

Dr. U Ravi Babu, IJECS Volume 6 Issue 1 Jan., 2017 Page No.19965-19968 Page 19967
 DOI: 10.18535/ijecs/v6i1.20

5. CONCLUSIONS [2] Liu Lizhen, Song Wei, Li Chuchu, Wang Hanshi and Lu
Jingli, “A Novel Feature-based Method for Sentiment
In this paper we analyze sentiments of reviews related to the Analysis of Chinese Product Reviews,” Proceedings of
ICT Management, China Communication (ICTM-2014),
services provided by e-shopping companies. For this we select pp. 154-164, March 2014.
top five most popular e-shopping sites. We collect reviews [3] Jalaj S. Modha, Sandip J. Modha and Gayatri S. Pandi,
from online sources. Sentiment detection from social media “Automatic Sentiment Analysis for Unstructured Data,”
streams is a difficult task. We used different preprocessing vol. 3 issue 12, pp. 91-97, December 2013.
techniques for removing unwanted things from reviews. It has [4] Subhabrata Mukherjee and Pushpak Bhattacharyya,
“Feature Specific Sentiment Analysis for Product
been observed that the pre-processing of the data and sampling Reviews,” Dept. of Computer Science and Engineering,
procedure are greatly affecting the quality of detected IIT Bombay, 2011.
sentiments. We find the sentiments for the five large e- [5] M. Thelwall, K. Buckley, G. Paltoglou, D. Cai, and A.
shopping dataset. It uses “Sentiwordnet dictionary” for finding Kappas, Sentiment strength detection in short informal
text, J. Amer. Soc. Inform. Sci. Technol., Vol. 61, No. 12,
score of each words. Then sentiments are classified as positive, pp. 2544–2558, 2010.
negative and neutral. This gives the best performance, thus [6] Y. Tausczik and J. Pennebaker, The psychological
being more reliable. The analysis of the services according to meaning of words: Liwc and computerized text analysis
positive, negative and neutral reviews can be represented methods, J. Lang. Soc. Psychol., vol. 29, no. 1, pp. 24–54,
2010.
graphically in experimental results section. These results can
[7] B. Pang and L. Lee, Opinion mining and sentiment
guide us to select particular site for e-shopping, based on analysis, Found. Trends Inform. Retrieval, vol. 2, no. (1–
maximum number of positive reviews. 2), pp. 1–135, 2008.
[8] Amit Pimpalkar, Tejashree Wandhe, Minal Kene and
Swati Rao, “Review of Online Product using Rule Based
References and Fuzzy Logic with Smiley‟s,” International Journal of
[1] Shulong Tan, Yang Li, Huan Sun, Ziyu Guan, Xifeng Yan, computing and technology (IJCAT), vol. 1, issue 1, pp. 39-
Jiajun Bu, Chun Chen and Xiaofei He, Interpreting the 44, February 2014.
Public Sentiment Variations on Twitter, IEEE
Transactions On Knowledge And Data Engineering, Vol.
6, No. 5, May 2013.

Dr. U Ravi Babu, IJECS Volume 6 Issue 1 Jan., 2017 Page No.19965-19968 Page 19968