0% found this document useful (0 votes)
18 views5 pages

Sentiment Analysis On Product Reviews-1

This paper presents a literature survey on sentiment analysis techniques applied to product reviews, focusing on methods to classify sentiments as positive, negative, or neutral. It discusses various approaches, including machine learning and lexical-based techniques, and highlights the effectiveness of algorithms like Support Vector Machine (SVM) and Naïve Bayes in achieving high precision in sentiment classification. The paper concludes that combining different techniques can enhance the accuracy of sentiment analysis in online reviews.

Uploaded by

gooohooohooo534
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
18 views5 pages

Sentiment Analysis On Product Reviews-1

This paper presents a literature survey on sentiment analysis techniques applied to product reviews, focusing on methods to classify sentiments as positive, negative, or neutral. It discusses various approaches, including machine learning and lexical-based techniques, and highlights the effectiveness of algorithms like Support Vector Machine (SVM) and Naïve Bayes in achieving high precision in sentiment classification. The paper concludes that combining different techniques can enhance the accuracy of sentiment analysis in online reviews.

Uploaded by

gooohooohooo534
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 5

2019 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS)

Sentiment Analysis on Product Reviews


Rahul Vasundhara Raj Monika
Department of Computer Science and Department of Computer Science and Department of Computer Science and
Engineering Engineering Engineering
Delhi Technological University Delhi Technological University Delhi Technological University
Delhi, 110042, India Delhi, 110042, India Delhi, 110042, India
[email protected] [email protected] [email protected]

Abstract— Sentiment Analysis is a process of analyzing and and Aggregation Scheme, Intrinsic and Extrinsic
categorizing the emotion or sentiment over any given review Domain Relevance Approach, etc can be used to extract
or text piece in order to know what the reviewer wants to the feature of the pre-processed text. After that
express in the form of positive, negative or neutral. Today, classification techniques are used to classify the
people are highly interested in buying things online from any sentiments into positive, negative or neutral followed by
e-commerce site or they search for a product review in order
to know the quality and one’s perception toward that product
polarity detection. These sentiments are determined over
before buying. The product provider also gets to know about various words, phrases. Final step is to verify the
the user’s opinion over a product. This can help the company precision of the result obtained.
to improve its marketing strategy and quality of product in
their favor. Sentiment analysis uses various semantic Different techniques used in classification of sentiments
approaches like on these online reviews to extract as much on different types of datasets from different platforms.
feature it can and categorize the type of opinion .Some These techniques help to detect polarity of the text or
techniques also help in rating the product value based on review used. Sentiment Analysis techniques includes
user’s opinion. This paper is a literature survey including
machine learning and lexical based approach. This paper
various authors and their sentiment techniques on product or
online review. includes collection of many research papers briefing
their techniques used in sentiment analysis classification,
Keywords— Sentiment Analysis, Online Reviews, Opinion their observation on their techniques used. Techniques
Extraction, Polarity, SentiWordNet, etc. like Support Vector Machine, Naïve Bayes, Neural
Network, etc. are highly used in this area. There are
I. INTRODUCTION many more techniques that work good on their datasets.
Online buying today is the most preferred and
trending way for buying any product like clothes, home The rest of the paper is structured as follows. Section
essentials, electronic gadgets, sport equipments, books, II introduces about sentiment analysis process model and
etc. For buying a particular thing, a user needs to be sure techniques used in it. Section III describes the various
about the quality of the product, as only product pictures techniques used by different authors to accomplish the
and description can be viewed online. So there is a classification of sentiment and perform these techniques
review system for the product, reviewed by the users on different review sets. In section IV finally, we
who bought it. They write about the product quality they conclude the paper.
received and other opinions over that product bought by
the. This review helps other user buying the similar II. SENTIMENT ANALYSIS PROCESS
product. And these review sets can be taken from MODEL AND TECHNIQUES
different platforms such as Amazon, Flipkart, Ebay,
taobao, yelp, etc. This Section of paper consists of Sentiment Analysis
Suppose a review says, “This product is awesome. Good Process Model and different approaches used in it to
to use.” It can be seen that the words like awesome and analyze the sentiment.
good has been used. These words indicate a positive
opinion. But to mine the sentiments technically these A. Process Model of Sentiment Analysis
reviews undergo assorted process and classification
techniques. Firstly, these reviews are gathered and Sentiment Analysis is a process to determine the
proper noise removal is done in order to process the person’s opinion or attitude over any product, service or
texts. Then different extraction techniques like Total organization. We will discuss each step in brief as
Weighted Score Computing Method, depicted in Fig 1.
Neutral/Polar/Irrelevant Classification Model, Weighing

ISBN: 978-1-7281-4826-7/19/$31.00 ©2019 IEEE 5


Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on January 24,2025 at 08:06:11 UTC from IEEE Xplore. Restrictions apply.
2019 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS)

4. Sentiment Classification: Now different sentiment


Data Text Pre-
Feature analysis techniques are used to classify the extracted
Selection adn
Acquisition processing
Extraction feature into various sentiments. Some of the techniques
widely used are Support Vector Machine (SVM), Naïve
Bayes, Neural Network, etc.
Validation and Polarity Sentiment 5. Polarity Detection: After the sentiment classification
Evaluation Detection Classification
step, polarity detection is done to determine the polarity
of statement, whether it is positive, negative or neutral.

6. Validation and Evaluation: Finally, validation and


evaluation of the obtained result is done in order to testify
Fig 1. Sentiment Analysis Process Model the precision of the result.
1. Data Acquisition: The first step of Sentiment Analysis B. Sentiment Analysis Technique
is to acquire or collect data on which sentiment
approaches has to be performed. These dataset can be any Sentiment Analysis uses different techniques for the
online review on any product from different sites or any classification of the extracted feature. These are classified
opinion. Without data collection model cannot step into Machine Learning Approach and Lexical Based
further. On these datasets analysis and classification of Approach. In section III, there is list of many works done
text can be performed. in the field of sentiment analysis techniques on product
review or online review. Machine Learning approach
2. Text Pre-processing: After data acquisition, data needs includes various other approaches under Unsupervised
to be pre-processed which includes removal of unwanted and Supervised learning with Support Vector Machine,
noise, emoticons used, duplicate words, stemming, Naïve Bayes, Decision Tree Classifier and Rule Based
different types of URLs, etc. to make data noise-free and Classifier. These algorithms work on sample datasets to
extract best features. determine prognosis or decision. The learning algorithms
differ in their approach and depend on input and output
3. Feature Selection and Extraction: This step is very data. Lexical Based approach includes Dictionary-based
important in order to get best out of nothing. Filtered data and Corpus-based technique. Fig 2. depicts the hierarchy
is selected and good extraction technique is used to of Sentiment analysis techniques. The Dictionary based
extract feature for the precision of the model. Different analysis uses the terms that are collected manually. This
extraction technique used in sentiment analysis is Total set grows by dictionary which contains similar or opposite
Weighted Score Computing Method, words. One example is WordNet. The Corpus-based
Neutral/Polar/Irrelevant Classification Model, Weighing approach works on providing dictionaries for a specific
and Aggregation Scheme, Intrinsic and Extrinsic Domain domain and further grows through search of similar words
Relevance Approach, etc. by using either of semantic technique or statistical
technique.

Sentiment Analysis

Machine Learning Approach Lexical- based Approach

Unsupervised
Supervised Learning Dictionary-Based Approach Corpus-Based Approach
Learning

Decision Tree
Linear Classifiers Rule-Based Classifiers Probabilistic Classifiers Statistical
Classifiers

Neural Bayesian Network Semantic


Network

Support Vector Machine(SVM) Naive Bayes

Maximum Entropy

Fig 2. Sentiment Analysis Approach

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on January 24,2025 at 08:06:11 UTC from IEEE Xplore. Restrictions apply.
2019 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS)

III REVIEW AND DISCUSSIONS

Need of product review has come in seen due to various precision.SVM is considered an adept technique as it gets
reasons. These reasons are like many companies not being maximum precision out of all three classifier techniques
tuned with requirement of customers, lack of a single and used. Moreover in future the datasets can be modified into
consolidated review for various products, assimilating the images from different social networking data and proper
varied thoughts to get a deterministic result is tough. So comparison can be done. Also for datasets used in [12]
the reviews should be authentic and genuine. Many resulted Naive Bayes to be better than other two methods
authors have researched and implemented different for text classification of opinion mining. Zeenia Singla et
approaches to classify the sentiment out of these reviews al.[13] used Statistical and Sentiment Analysis
with high accuracy. framework, SVM for Cross-validation on top three brands
of mobile- Samsung, BLU and Apple. Through statistical
There are many papers analyzing and implementing analysis, Samsung gets most positive feedback. High
various sentiment approaches to extract the feature and priced product had better rating, better quality and
classify the polarity of given datasets (reviews). customer satisfaction. These brands had positive reviews
Alexandra Cernian et al. [1] used SentiWordNet lexical with positive sentiments of trust, joy, anticipation and
resource on set of 300 product reviews from Amazon.This surprise by sentiment orientation. SVM applied on cross
technique proved 61% success rate. The result produced validation gave 84.87% precision on classified data. In
is distinguished by the star ratings in Amazon to obtain a future, this system can be made more focused for the
depiction of precision of approach used. Further an benefit of the industry. To mine user’s review and
optimized algorithm can be planned to extend the text requirements and provide useful information enabling the
source semantic processing. Feature based sentiment designer to improvise the product and make requirement
classification approach with Tree Bank Model & Fuzzy acquisition less time consuming so that less expense is
Opinion Mining is used by K Indhuja et al. [2] on user incurred in surveys, interviews, market research, etc.
spawned product reviews from different websites. The
anticipated system works with more competence of Anh-Dung Vo et al. [9] proposed extraction and
85.58% precision. The projected model for imitating summarization of the facets of the product from large
fuzzy linguistic hedges is effectively applied into the number of reviews of different product domain using
sentiment categorization task.This system can be further Two stage systems - First is Knowledge extraction using
broaden to rate the authenticity of a review based on the NLP tools (DP, CR, NER) for the extraction of syntactic
opinion of other reviewers. Neuro-Fuzzy system may lead knowledge and entail opinion-facet connection. Second
to new advancement. stage is Sentiment Analysis on knowledge extracted from
first stage.For reviews on the product camera, the
P.V. Rajeev et al.[4] and Esha Tyagi et al. [5] used Naive projected system attains F1-scores of 0.714 and 0.774 for
Bayes and SVM for Feature extraction of reviews laptop reviews. For the product of laptop, the facet
(product) and accredit the respective polarity on online extraction attained satisfactory results. This approach can
reviews for smart phone products as dataset. The be used to extract facets of product on large datasets and
proposed system [4] gave score classification in star based can be applied to tough challenges like sarcastic
rating and helps in picking the appropriate product.The statements and implicit opinion inferences.
proposed system used only one product at a time. So in
future, the technique can be modified in order to use more Shaozhong Zhang et al. [10], proposed different trust
than one product and automatically rank the product computation algorithm to find trust between two users
according to their feature. Whereas, [5] used support seeking a product review from e-commerce site using
vector machine which found to be 89.98% accurate on Sentiment Resemblance Analysis. For sentiment
given datasets. Some papers do have comparison of resemblance computation- fine grained analysis method is
different classification techniques. Marium Nafees et used, for direct trust computation- weighted average
al.[11], Santosh Kumar KL et al.[12] both used Naive method is used and for propagation trust computation-
Bayes, SVM and Logistic Regression,Sentiwordnet algo graph based propagation algorithm is used on reviews
on reviews from twitter and Amazon respectively. The from different category (Books, electronics, Sports, Video
techniques used [11] are good to conclude Sentiment games, Baby) from Amazon.com. For direct trust
analysis, but SVM is best approach to attain utmost relationship the precision value should be set to 1.5. For
precision of polarity division. In Cumulative analysis propagation trust experiment, threshold value is used that
result, SVM gets overall precision of 71.80%, Naïve indicates trust path length between two user or node.
Bayes has 56.4% and Logistic Regression has 62.4% These values are important in improving the propagation

7
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on January 24,2025 at 08:06:11 UTC from IEEE Xplore. Restrictions apply.
2019 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS)

trust link analysis. But it is really challenging in real E- Yan Wan et al.[6] proposed Fine-grained
commerce system. Sentiment Analysis with inherent feature
based on POS rules and make a niche
Yan Wan el al.[6], proposed Fine- grained Sentiment market analysis to help seller and buyer.
Analysis with inherent feature based on POS rules on
Product reviews and make a niche market analysis to help
manufacturers and also for consumer on online reviews of In [7], Xiaojiang Lei et al. proposes a
one dress from taobao.com site. Precision of fine-grained 2016 sentiment-based rating prognosis method
sentiment analysis is improved by uniting implicit (RPS) to improve the precision of prognosis
features. Generality of analyzed sentiment potency is also in recommender system.
increased. This system helps manufacturers to know what
to improve in their products and discover niche market. [9], [11] classify the polarity of reviews
Also customers get to know the advantages and 2018 using techniques like NLP, SVM, Naïve
limitations of the product they want to buy and can make Bayes, Logistic Regression on online
good decision. So, the system improves the precision to datasets.
analysis.
Shaozhong Zhang et al. [10] aimed to find
XIA et al. [3] dealt with the polarity shift problem in
trust between two users seeking a product
sentiment classification for online reviews using Dual 2019
review from e-commerce site using
Sentiment Analysis(DSA), DSA-WN, DSA-MI. DSA Sentiment Resemblance Analysis.
model plays an effective role for polarity classification
and it stands ahead of assorted substitute methods for
polarity shift. This system creates conversed reviews to
abetment supervised sentiment classification. So, it can be IV CONCLUSION
generalized to wider range of SA task and focus on
considering complex polarity shift patterns(transitional, This paper includes various papers supporting
sentiment-inconsistent & subjunctive) to create conversed different techniques applied on online reviews in order to
review. In [7] Xiaojiang Lei et al.'s work a recommender find polarity or to classify those review into various
system is used with the fusion of user sentiment sentiments. Support Vector Machine and Naïve Bayes
resemblance, actuating sentiment between persons, and technique has provided good and adept results on their
item prominence resemblance into a single matrix respective datasets. SVM also gives best results for
factorization framework. It is used to improve the computation of polarity of the ambiguous sentiment or
prognosis of ratings for a product. A new relationship is reviews. For polarity shift problem Dual Sentiment
build between user and friend called interpersonal Analysis is effective. The Part of Speech (POS) rule
sentiment influence. It reflects the sentiment angle to user increases the sentiment potency. So combining POS and
as per the influence of user’s friend.The three sentimental SVM is effective. SentiWordNet analysis can also be
factors that are participating make great contribution to looked up for classifying reviews. To encounter user
prognosis of rating. This system shows significant sentiment resemblance, fine grained analysis can be used.
improvement over extant approaches.
Future work can be seen by applying techniques using
Table I. Research in Classification of Sentiment on image kind of data as dataset. Above paper lacks the
product reviews using various techniques. analysis on sparse data i.e. if user has few reviews. It
would be challenging to work on sarcastic statements and
Year Contributions opinions which are not directly expressed or implicit
opinion inferences.
Feature based sentiment classification
2014 approach with Tree Bank Model & Fuzzy
Opinion Mining is used by K Indhuja et al.
REFERENCES
[2]

Alexandra Cernian et al. [1] used [1] Alexandra Cernian, Valentin Sgarciu, Bogdan Martin:
SentiWordNet lexical resource on product Sentiment analysis from product reviews using
2015 reviews. SentiWordNet as lexical resource, 978-1-4673-6647-
[3] To deal with the polarity shift problem 2/15/$31.00 ©2015 IEEE
[2] K Indhuja, Raj P C Reghu. “Fuzzy Logic Based Sentiment
in sentiment classification for online Analysis of Product Review Documents”, IEEE First
reviews using Dual Sentiment Analysis. International Conference on Computational Systems and

8
Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on January 24,2025 at 08:06:11 UTC from IEEE Xplore. Restrictions apply.
2019 International Conference on Computing, Communication, and Intelligent Systems (ICCCIS)

Communications, 2014. [11] Marium Nafees et al. : Sentiment Analysis of Polarity in


[3] Rui Xia, Feng Xu, Chengqing Zong, Qianmu Li, Yong Qi, Product Reviews In Social Media ,
and Tao Li: Dual Sentiment Analysis: Considering Two 10.1109/ICET.2018.8603585
Sides of One Review, VOL. 27, NO. 8, AUGUST 2015, [12] Santhosh Kumar K L, Jayanti Desai, Jharna Majumdar:
1041-4347 _ 2015 IEEE Opinion Mining and Sentiment Analysis on Online Customer
[4] P.Venkata Rajeev, V.SmrithiRekha “Recommending Review: 978-1-5090-0612-0/16/$31.00 ©2016 IEEE
Products to Customers using Opinion Mining of Online [13] Zeenia Singla, Sukhchandan Randhawa, and Sushma Jain:
Product Reviews and Features”,International Conference on STATISTICAL AND SENTIMENT ANALYSIS OF
Circuit, Power and computing technologies,2015 CONSUMER PRODUCT REVIEWS:
[5] Esha Tyagi and Arvind Kumar Sharma: 10.1109/I2C2.2017.8321910
Sentiment Analysis of Product Reviews using Support Vector [14] JINMING ZHANG1, DEXIN CHEN2, AND MIN LU 1:
Machine Learning Algorithm Vol 10(35), Combining Sentiment Analysis With a Fuzzy Kano Model
DOI:10.17485/ijst/2017/v10i35/118965, for Product Aspect Preference Recommendation,
September 2017 10.1109/ACCESS.2018.2875026 VOLUME 6, 2018
[6] Yan Wan, Hongzhurui Nie, Tianguang Lan, Zhaohui Wang: [15] Xu Yuan, Mingyang Sun, Zhikui Chen , (Member, Ieee), Jing
Fine-grained Sentiment Analysis Of Online Reviews: 978-1- Gao, And Peng Li: Semantic Clustering-Based Deep
4673-7682-2/15/$31.00 ©2015 IEEE Hypergraph Model for Online Reviews Semantic
[7] Xiaojiang Lei, Xueming Qian and Guoshuai Zhao: Rating Classification in Cyber-Physical-Social Systems,
Prediction Based on Social 10.1109/ACCESS.2018.2813419 2169-3536 2018 IEEE,
Sentiment from Textual Reviews: 1520-9210 © 2016 IEEE. VOLUME 6,2018
[8] Wei Zhao , Ziyu Guan , Long Chen , Xiaofei He, Deng Cai , [16] Gurneet Kaur, Abhinash Singla: Sentimental Analysis of
Beidou Wang , and Quan Wang: Weakly-Supervised Deep Flipkart reviews using,Naïve Bayes and Decision Tree
Embedding for Product Review Sentiment Analysis: 1041- algorithm, International Journal of Advanced Research in
4347 _ 2017 IEEE Computer Engineering & Technology (IJARCET), Volume 5
[9] Anh-Dung Vo , Quang-Phuoc Nguyen , And Cheol-Young Issue 1, January 2016.
Ock: Opinion_Aspect Relations In Cognizing Customer [17] Pankaj Gupta, Ritu Tiwari and Nirmal Robert: Sentiment
Feelings Via Reviews: 2169-3536 2018 IEEE Analysis and Text Summarization of Online Reviews: A
[10] S. Zhang, H. Zhong: Mining Users Trust From E-Commerce Survey, 978-1-5090-0396-9/16/$31.00 ©2016 IEEE
Reviews Based on Sentiment Resemblance Analysis,
VOLUME 7, 2019 2169-3536 2019 IEEE.

Authorized licensed use limited to: ANNA UNIVERSITY. Downloaded on January 24,2025 at 08:06:11 UTC from IEEE Xplore. Restrictions apply.

You might also like