International Conference on Computing, Communication and Automation (ICCCA2017)

SENTIMENT ANALYSIS ON PRODUCT

REVIEWS
ChhayaChauhan*1, Research Scholar SmritiSehgal2, Assistant Professor
Department of Computer Science and Engineering Department of Computer Science and Engineering
Amity University, Uttar Pradesh, India. Amity University, Uttar Pradesh, India.
[email protected] [email protected]

Abstract—Sentiment analysis is used for Natural language understanding such sentiments. Firstly, data is collection from
Processing, text analysis, text preprocessing, Stemming etc. are the e-commerce store amazon.com and Flipkart.com. It
the major research field in current time. Sentiment analysis using consists of customer reviews about various products such as
different techniques and tools for analyze the unstructured data mobile phones, television, camera etc. However, for our
in a manner that objective results can be generated from them. project, the focus is on electronic goods. These include but are
Basically, these techniques allow a computer to understand what not limited to phones, televisions, laptops and cameras. The
is being said by humans. Sentiment analysis uses different data is collected dynamically through web scraping. Then, the
techniques to determine the sentiment of a text or sentence. The polarity of the opinions must be analyzed. For instance,
Internet is a large repository of natural language. People share “good” and “bad” respectively hold a negative and a positive
their thoughts and experiences which are subjective in nature. opinion. Secondly, for more accurate result can be obtained,
Many a time, getting suitable information about a product can the opinion strength of the opinion should determine. If the
became tedious for customers. Companies may not be fully aware strength of words used to present the user’s opinion is taken
of customer requirements. Product reviews can be analyzed to into account. For example, “good” and “excellent” indicate
understand the sentiment of the people towards a particular
different levels of positive sentiment. The addition of “very”
topic. However, these are voluminous; therefore a summary of
before any word can also be used to determine the strength.
Finally, the classification of review is classified with respect
positive and negative reviews needs to be generated. In this
to sentiment classes, i.e. the polarity of reviews are deciding
paper, the main focus is on the review of algorithms and
which are positive and negative reviews.
techniques used for extract feature wise summary of the product
and analyzed them to form an authentic review. Future work will
The main objective of sentiment analysis on product reviews
include more product reviews websites and will focus on higher is to review different algorithm and techniques to extract
level natural language processing tasks. Using best and new feature wise summary of a product and analyze it to form an
techniques or tool for more accurate result inwhich the system authentic review. Sentiment analysis for a particular product
except only those keywords which are in dataset rest of the words can help companies know about these expectations before
are eliminated by the system. product launch. Thus, product analysis can prove beneficial to
companies and allow them to develop efficient marketing
Keywords—Sentiment analysis; polarity; Natural language strategies. The customers will be able to judge a product
processing; Product reviews; Senti-word net; Part-of-speech wisely before purchasing it. Comparative analysis will allow
tagging them to pick the right product and the right brand that suits
their requirements. The determination of features and their
rating will help customers take an informed decision.
I. INTRODUCTION
The aim of sentiment analysis on product review are large
In Web 2.0, Many Customers buying or selecting product number of products are being talked about on the internet. The
according to reviews on Internet. The internet is a large problem is to extract the information for a particular product
repository of natural language. People share their thoughts and or brand and classify the opinion based on various product
experiences which are subjective in nature. Many a time, features. This involves extracting features and training the
getting suitable information about a product can became classifier based on the same. The features should be such that
tedious for customers. Companies may not be fully aware of the training speed is improved. A good accuracy on test data
customer requirements. Product reviews can be analyzed to has to be achieved. The features of a product that are being
understand the sentiment of the people towards a particular talked about have to be determined. That is, it should be
topic. The sentences below are extracted from a product known to the user’s the basic on which the product is given a
review on the Internet product feature database: “In Sony certain rating. The features have to be classified into positive
mobile, camera is very good but battery life is bad”. In this and negative.
example, reviewer stated that in Sony mobile, camera is very The need of sentiment analysis for product arises due to a lack
good but battery life is bad. Several tasks are involved to of a single, consolidated review for various products. A
ISBN:978-1-5090-6471-7/17/$31.00 ©2017 IEEE 26
 International Conference on Computing, Communication and Automation (ICCCA2017)

customer may come across positive and negative reviews and Sentiment analysis for a particular product can help companies
opinions, but assimilating these varied thoughts to get a know about these expectations before product launch. Thus,
deterministic result is tough. Through Sentiment analysis on product analysis can prove beneficial to companies and allow
product reviews, the customers will be able to judge a product them to develop efficient marketing strategies. The customers
wisely before purchasing it. The analysis will allow them to will be able to judge a product wisely before purchasing it.
pick the right product and the right brand that suits their Comparative analysis will allow them to pick the right product
requirements. The determination of features and their rating and the right brand that suits their requirements. The
will help customers make a smart choice. Many companies are determination of features and their rating will help customers
not in tune with requirement of the customers. They might be take an informed decision.
spending millions of dollars in the research and development The next section of this paper is arranged as follows:
of the product, but still fail to understand why their product is InLiterature Survey section provided a brief review of various
not creating the required buzz in the market. Use of product techniques implemented for sentiment analysis on product
analysis after product launch can help understand the flaws as reviews. In Conclusionsection concludes the paper. InFuture
well as the upper selling points of the product. Many a time, Worksection, include the future work.
products do not meet the expectations of the customers.

II. LITERATURE SURVEY

Table I. Various techniques implemented for sentiment analysis on products reviews

P. FOCUS METHODS/ SAMPLE RESULT ADVANTAGES PROBLEM FUTURE WORK/
NO. TECHNIQUE CONCLUSION

Focus is to . Feature . For

determine the basedSentim classification, The future work
subjectivity ent In this system does not In this paper have lots of
reviews, classification paper, 2000 required labeled main focus was challenges quite
polarity and approach user The accuracy of training set. to determine the exist in the field of
polarity was used. generated the system was subjectivity machine learning:
[ 1] strength of product 58.58% and reviews but Co- reference
reviews by
extracting
.There are reviews
were used
systemgave
good
.The system objectivity
reviews wasn’t
Resolution;
Domain
two models takes benefits of
features from were used: from performance both neural determined. Dependency; Entity
product Tree Bank different network and Recognition etc. ,
reviews and Model and websites. fuzzy network these problem will
classify the Fuzzy which may guide have to tackled and
reviews into Opinion to new in data the solutions of
positive, Mining mining field. these problems will
negative and Model. used to improve the
neutral class. sentiment analysis.

27
 International Conference on Computing, Communication and Automation (ICCCA2017)

Focus is
todetermine
. Feature In this The result gave .In this paper, The problems in The future work
wise opinion paper, data numeric score both features as this paper was will to improve the
the intensity mining collected of all features well as modifiers that the system recall values by
of the system was from e- and after that was identified. give low recall using best system
opinions by used. commerce calculating the values and due to for correct English
giving
attention to
sites i.e.
Amazon and
overall ordering
of feature to
. The weight of low recall values
sizeable amount
and correct POS
and the system in
[2] each feature was
modifier flipkart and determine how of reviewers do which parser
calculated and
words which dataset intense the not use correct successfully
based on score
expresses consists of opinion for English and does identify the
values assign the
opinions and 1000 positive not gave correct sentences.
rank of all
extract the objective features and POS as well as
features.
features from sentences negative the parser
user and 1000 features. unsuccessful to
generated subjective identify the
reviews. sentences. sentences.

Focus is to In this The result gave .The system The system used The future work
identify .Senti-Word paper, 5 higher precision finds higher Senti-word net will improve the
opinion Net product set and recall. precision and approach but this Senti-word
sentences in approach of reviews recall. approach doesn’t approach and
each review. was used. were used. sort features according to the
[3]
Pruning
features is
.The system was from strongest to
weakest order.
opinion strength
classifies the
used pruning
done to feature to features and sort
remove remove incorrect them from strongest
incorrect features. to weakest order.
features.

The result
Focus is on In this showed that
preprocessing .Feature paper, five after the The main
to incorrect based different compactness advantage of this
and unwanted sentiment products pruning, the paper was that The accuracy of The future work
words, classification from precision values the system used the system was will to find more
extraction of approach Amazon were increased two types of 79.67% which is features for opinion
features and was used. were used. and recall pruning i.e. less than Fuzzy mining and applies
[4] POS tagging values were compactness and Logic Based more classifiers for
for polarity. decreased in the p- support approach. the work.
reviews pruning to
whereas p- compare
support precision and
pruning, the recall values.
precision values
increased but
recall values
remains same.

28
 International Conference on Computing, Communication and Automation (ICCCA2017)

Focuses is to In this The result The system

analyzed and .Ontology- paper, showed optimal provides more
classifycusto based 20,000 hyper plane successful
mer’s product sentiment product curve i.e. sentiment - -
[5]
reviews. classification reviews decision classification
was used. were boundary based on
selected for between reviews.
trained the suitable and
sentiment unsuitable data.
classifiers
models.

Focus is to In this The result gave

classify the .WEKA paper, more positive, The system was
online classifier than 5000 negative and classified the The drawback of The future work
reviews. method was reviews neutral reviews reviews into WEKA tool was will to combine
used. were based on positive, that many some features of
[6] selected. polarity and classifiers of classifiers to solve
negative and
with the help of neutral class WEKA were the problem and
polarity of based on lacking in expanding hybrid
product reviews polarity. accuracy. classifier and
consumers will implement them.
decide to buy
the product or
not.

Focus is on The result .The The main

topic .Combined In this gavegood performance of drawback was
detection and Sentiment paper, the performance in CST model was that the system
sentiment Topic (CST) unstructured multiple higher than other doesn’t find
analysis of model was documents domains which approaches. neutral class. -
[7] reviews. The for express the
used.
system is
purely based
classificat-
ion were
pliability of
CST for
. The system
used CST model
on selected. sentiment to improve
unsupervised analysis work. accuracy and
learning efficiency.
technique.

Focus is to Naive Bayes In this The advantage The drawback The future work
extract the classification paper, only The system was that system was that the will to extract more
features of was used mobile gave score helps in choosing system used only than 2 products and
reviews and phone classification in the right product one product. improved the
[8] assign the reviews was star rating. techniques which
polarity to select from automatically rank
that features. flipkart sites the products based
on the features.

29
 International Conference on Computing, Communication and Automation (ICCCA2017)

III. CONCLUSION REFERENCES

It is concluded that Naïve Bayes’ classifier can gave very [1] K Indhuja, Raj P C Reghu. “Fuzzy Logic Based Sentiment
good results when good features were extracted. The results Analysis of Product Review Documents”, IEEE First
vary for all the n-grams and it has been observed that International Conference on Computational Systems and
unigrams and bigrams give the best results. Also, feature Communications, 2014.
presence was a better measure for sentiment analysis than
[2]Tanvir Ahmad, Mohammad NajmudDoja. “Ranking
feature frequency. The results for different type of techniques
System for Opinion Mining of Features from Review
vary depending on the kind of dataset used. POS tagging and
Documents”, International Journal of Computer Science
use of negation improves accuracy. Some techniques of
Issues, Vol. 9, Issue 4, No 1, July 2012.
unsupervised learning can also be taken into account. These
can be used for aspect level sentiment analysis. These include [3]Weishu Hu, Zhiguo Gong, JingzhiGuo. “Mining Product
finding the features and opinions using POS tagging and Features from Online Reviews”, 7th IEEE International
classifying the opinion based on its semantic orientation as Conference on E-Business Engineering, 2010.
determined by Word Net. The word presence gave better
result than word frequency. This result is consisting with [4]SiddharthAravindan and AsifEkbal.“Feature Extraction and
previous research work. The present to irrelevant sentences Opinion Mining in Online Product Reviews”, International
gave no output. The sentence which contains a feature and Conference on Information Technology, 2014.
opinion both gave a meaningful output. The specific feature
corresponding to opinion i.e. Octacore processor and LTE [5]JantimaPolpinij, Aditya K. Ghose. “An Ontology-based
SIM slot were also extracted. Feature wise summary was Sentiment Classification Methodology for Online Consumer
obtained. Reviews”, IEEE/WIC/ACM International Conference on
Web Intelligence and Intelligent Agent Technology,Vol. 1,
2008.
IV. FUTURE WORK
[6]Aditya A. Kshirsagar, Prarthana A. Deshkar. “Review
The future work is to use data from websites such as EBay Analyzer Analysis of Product reviews on WEKA
and twitter to aggregate their reviews and provide a larger Classifiers”,IEEE Sponsored 2nd International Conference on
number of numbers of reviews because in previous work, data Innovations in Information, Embedded and Communication
are extracted from Amazon and Flipkart. The various systems, 2015.
performance measures such as precision and recall can be
[7] M.S. Usha, M.Indra Devi. “Analysis of Sentiments using
incorporated to analyze the performance of the classifier. A
Unsupervised Learning Techniques”,IEEE International
more accurate result can be obtained if the strength of words
Conference on Information Communication and Embedded
used to present the user’s opinion is taken into account. For
Systems, 2013.
example, “good” and “excellent” indicate different levels of
positive sentiment. The addition of “very” before any word is [8]P.Venkata Rajeev, V.SmrithiRekha “Recommending
use to determine the strength. Future work will include more Products to Customers using Opinion Mining of Online
product reviews websites and focus on higher level natural Product Reviews and Features”,International Conference on
language processing tasks. Using best and new techniques or Circuit, Power and Computing Technologies, 2015.
tool for more accurate result in which the system except only
those keywords which are in dataset rest of the words are [9] Alexandra cernian ,ValentinSgarciu and Bogdan Martin.
eliminated by the system. “Sentiment analysis from product reviews using
SentiWordNet as lexical resource”, IEEE 7th International
Conference on Electronics, Computers and Artificial
Intelligence (ECAI), 2015.
ACKNOWLEDGMENT
[10]PankajGupta,RituTiwari and Nirmal Robert.“Sentiment
I thank my college Amity University, Noida for giving us analysis and text summarization of online reviews: A survey”,
an opportunity to be a part of summer internship that leads to IEEE International Conference on Communication and Signal
the development of this work. Many thanks to Processing (ICCSP), 2016.
Ms.SmritiSehgalfor countless discussions and feedback that
helped me to complete the worksuccessfully.

30
 International Conference on Computing, Communication and Automation (ICCCA2017)

[11]PingJi and Jian Jin. “Extraction of comparative opinionate [13] I. Hemalatha,G.P.SaradhiVarma and A.
sentences from product online reviews”,12th International Govardhan.“Sentiment classification in online reviews using
Conference on Fuzzy Systems and Knowledge Discovery FRN algorithm”, IET Chennai Fourth International
(FSKD), 2015. Conference on Sustainable Energy and Intelligent Systems
(SEISCON), 2013.
[12]M.LovelinPonn Felciah and R.Anbuselvi. “A study on
sentiment analysis of social media reviews”, IEEE [14]R.Nithya and D.Maheswari. “Sentiment Analysis on
International Conference on innovations in Information, Unstructured Review”, International Conference on Intelligent
Embedded and Communication Systems (ICIIECS), 2015. Computing Applications (ICICA), 2014.

31