Sentiment Analysis on Online Product Reviews

Dr. Rajesh Bose1, Raktim Kumar Dey2, Sandip Roy3, *, Dr. Debabrata Sarddar4
1, 2
Simplex Infrastructures Ltd., 3 Brainware Group of Institutions, 4 University of Kalyani
Kolkata, India.
{bose.raj00028, deyraktim, *sandiproy86, dsarddar1}@gmail.com

Abstract. Today people are exchanging their thoughts through online Web
forums, blogs, and different social media platforms. Sometime they are giving
reviews and opinions on different products, brand, and their services. Their
reviews towards a product is not only improves the product quality but also
influence purchase decisions of the consumers. Thus, product review analysis is
widely accepted platform where consumer can easily aware about their
requirements. In this experiment, we track 568,454 fine food reviews of 74,258
products and 256,059 users on Amazon over a period of ten years. To analyze
the result, we select six most popular products and users based on the plain text
review and NRC emotion lexicon is used which can be categorized eight basic
emotions and two sentiments. Wordclouds also help our research to make
comparisons between the eight emotion categories. Our results show that how
sentiment analysis will help to identify the consumers’ behaviors and overcome
those risks to meet the consumers’ satisfaction.

Keywords: Amazon’s customer reviews, Digital Marketing, Electronic word-

of-mouth (e-WOM), Sentiment Analysis, Unstructured data.

1 Introduction

Today, electronic word-of-mouth (e-WOM) is one of the most important factors for
digital marketing [1]. Companies are using the digital platforms for promoting their
products. Nowadays customer’s online reviews can influence the purchase decisions
of a product. Though, Amazon, eBay, Flipkart, and Walmart, different e-commerce
companies are analyzing their customers’ reviews in different ways [2, 3].
Unstructured data such as text, audio, photos, or videos are posted by the
customers in social media websites, Web forums, blogs, and different online review
platform [4, 5]. Natural Language Processing (NLP) is an automatic programming
technique to analyze and comprehend large amounts of customers’ opinions where
different companies build a chatbots to support online customer service interactions
[6, 7, 8].
Sentiment classifications play an important role for classifying unstructured data.
Cui et al classified online product reviews into positive and negative classes [9]. In
their experiment, they used different machine learning algorithms to evaluate different
trade-offs of 100K of online-product reviews. Fang et al tackled the sentiment polarity
categorization problem in online product reviews of Amazon.com [10]. They
categorized their outcomes into two level; a) sentence-level categorization, b) review-
level categorization.
In this manuscript, we used NRC emotion lexicon that categorized customers’
reviews into eight emotions (anger, fear, trust, anticipation, sadness, surprise, disgust,
and joy) and two sentiments (positive and negative) [11]. Our proposed analyzing
system can enhance the review technique, which has discussed by Escalona [12].
The detailed sections of the manuscript are as follows. Section 2 is described
literature Survey. Section 3 deals with on data preparation along with review analysis.
Result analysis part is discussed in section 4 and the section 5 draws conclusions and
examines the possibilities of next future.

2 Literature Survey

Information can be classified into two basic statements; a) Objective statements, b)

Subjective statements. Subjective statements are the opinions that consider the
people’s sentiments about entities and events [13]. Bhatt et al proposed a system that
classified the customer reviews along with the sentiment of the reviews [13].
Patel proposed a machine learning technique that classifies sentiments of review
based on words i.e. positive or negative [14]. Sygkounas et al tested his work on the
SemEval Twitter-based corpora and ESWC semantic sentiment analysis challenge
which get the result 88.05% accuracy over the test set for detecting the positive and
negative sentiment [15].
In [16], authors shown that Support vector machine (SVM) perform better
accuracy compared to Naïve bayes and maximum entropy methods. Leung et al also
classified the sentiment into two classes such as Positive and Negative [17].
Ray et al proposed a framework for sentiment analysis using R software. In their
method, they collected the plain text from Twitter, pre-processed, and then lexicon
based approach is used for analyzing user’s sentiment [18].
Govindaraj [19] proposed a method which extracts two sets of features from the
customer reviews; acoustic features which represent emotions and lexical features that
represent sentiments. These sets used in a supervised classifier to predict the
sentiment of the customers. In their experiments, they used an audio speech dataset
prepared and downloaded from Amazon and YouTube online portal respectively.
In the experiment [20], author designed a text classifier using Python Data
Analysis Library, Natural Language Toolkit and Scikit-learn libraries. McAuley et al
tested that open-ended queries performed better than binary queries [21].
Babu proposed a sentiment analysis technique for e-shopping websites [22]. He
used SentiWordNet dictionary to find scores of each word [23, 24]. Kaur et al
presented a combined approach of SVM and K-nearest neighbor classification to
review of the technical article [25]. Li proposed a framework to identify two stages:
feature extraction and classifier learning [26, 27].
Bose et al proposed the emotion analysis using syuzhet in CRAN package which is
based on NRC emotion lexicon [28]. In our proposed method, we used the same
package to categorize emotion into eight sub-categories and two sentiments [28].
3 Data Preparation

Dataset has collected from the Kaggle which consists of reviews of fine foods from
Amazon from October 1999 to October 2012 [29, 30]. Dataset includes 568,454
reviews, 74,258 products, 256,059 users, and 260 users with more than 50 reviews.

3.1 Data Pre-processing

Dataset in the form of plain text review is mainly unstructured data [31]. At first, we
apply data pre-processing method on our dataset [32, 33]. In this data preprocessing
steps, we have followed the steps below [34]:
 Remove all URLs (e.g. www.abc.com), screen name (e.g. @username), and all
hashtags (e.g. #topic)
 Remove all symbols, punctuations, and numbers
 Remove Stopwords
 Substitute any non UTF-8 by space
 Replace all the emoticons with their sentiment
 Change text to lowercase
 Replace words with their stems or roots
 Remove the retweets

3.2 Product Review Analysis

In our experiment, NRC emotion lexicon, large word list (like other lexicon AFINN,
ANEW, EmoLex, LabMT, General Inquirer, and SentiWordNet), is used for
sentiment analysis [11, 35]. This lexicon (version 0.92) had about 14,200 unigrams
word types with word-level emotion [11, 36].
Fig. 1 and 3, bar charts represent the eight emotions (anger, fear, trust, anticipation,
sadness, surprise, disgust, and joy) and two sentiments (positive and negative). We
track six most popular users and products for our sentiment analysis.
For user-centric approach, we found potential reviewer who made their own
opinion to the m number of products on the Amazon. Our aim is to find out the
sentiment based on their reviews. These analyses will help to identify the sentiment of
the customers. Similarly, in product-centric approach, we found the best reviewed-
product which has reviewed by n number of customers. We also analyzed the
sentiment or emotion of the n number of customers towards the products.
Fig. 2 and 4, word cloud is used to visualize review data which illustrates the
keywords used more frequently by the customers. We categorized our reviewed text
into eight emotion categories [37, 38].
In each category, highest frequency word will identify the most insightful of the
customers [39]. In our experiment, most of the amazon customers reviewed by the
below keywords; e.g. “good” or “wonderful” in surprise section and “flavor” or
“organic” or “smell” in disgust section.
Fig. 1. Distribution of emotion of six most popular product reviews of Amazon
users.
Fig. 2. Word Cloud of six most popular product reviews of Amazon users.
Fig. 3. Distribution of emotion of six most popular product reviews of Amazon
products.
Fig. 4. Word Cloud of six most popular product reviews of Amazon products
4 Result Analysis

In this manuscript, we analyzed the sentiment or emotion of the products based on

customers reviews of Amazon.com. Our aim of this research is to identify the highest
reviewed product and also track the most anxious word in each emotion category of
the customers using word cloud.
In the fig. 1 and 3, most of the popular product reviews of Amazon users and
Amazon products respectively. Most of the reviews have more polarity for “positive”,
“joy”, and “trust” rather than “fear”, “anger”, and “disgust”.
Differently, in fig 5 shows that “negative”, “disgust” and “anger” have more
polarity than “trust”, “joy” and “surprise” for the sentiment analysis of the least
popular product reviews in Amazon.com.
Furthermore, authority of the amazon product has more concern about those words
e.g. “vomit”, “smells” and “rotten” that users are more conscious about the quality of
the product.

Fig. 5. Sentiment analysis of least popular product reviews of Amazon product

5 Conclusion and Future Work

In our research, we have done the sentiment or emotion classification using NRC
emotion lexicon [11, 36]. Our experiment results show that how customers’ reviews
are important for digital marketing research [40, 41]. Sometime customers given
ratings differ with their comment [42, 43, 44]. Researchers work on the sentiment
analysis of the customers’ reviews of different e-commerce websites [45, 46].
Our research is not free from constraints. Future work may be performed as below:
we will include some topic modeling based sentiment analysis features on the plain
text reviews of the customers on e-commerce websites, which will predict the best
product as per the customers’ needs [47]. We will also perform perplexity analysis
and lexicon quality assessment on plain text review data of the customers in the near
future [48].
References

1. Wang, W., Wang, H., Song, Y.: Ranking product aspects through sentiment analysis of
online reviews. J. Exp. Theo. AI. 29, 227 – 246 (2017)
2. Waldron, M.: Building a Text Analysis process for customer reviews in RapidMiner.
AYLIEN (2016)
3. Spiewanowaski, P.: How to Visualize and Analyze Customer Feedback. uxBOOTH (2016)
4. Aswani, S.: Analyzing Customer Feedback Data: Manual Analysis vs NLP.
CLARABRIDGE (2017)
5. Luminoso.: Customer Feedback and Artificial Intelligence. White Paper (2016)
6. Orf, D.: Google Assistant Is a Mega AI Bot That Wants To Be Absolutely Everywhere.
Google (2016)
7. Thematic.: Visualizing Customer Feedback: Three Alternatives to Word Clouds. Medium
(2017)
8. Abuleil, S., Alsamara, K.: Using NLP Approach for analyzing Customer Reviews. SOEN –
2017, 117 – 124 (2017)
9. Cui, H., Mittal, V., Datar, M.: Comparative Experiments on Sentiment Classification for
Online Product Reviews. American Association for AI, 1265 - 1270 (2006)
10.Fang, X., Zhan, J.: Sentiment analysis using product review data. J. Big Data. 2:5, 1 – 14
(2015)
11.Mohammad, S.M., Turney, P.D.: Crowdsourcing a Word-Emotion Association Lexicon.
National Research Council, Canada, 1 - 25 (2013)
12.Escalona, T.: Detect sentiment from customer reviews using Amazon Comprehend. Amazon
Machine Learning Blog (2018)
13.Bhatt, A. Patel, A. Chheda, H. Gawande, K.: Amazon Review Classification and Sentiment
Analysis. IJCSIT. 6, 5107 – 5110 (2015)
14.Patel, B.: Predicting Amazon product reviews’ ratings. Medium (2017)
15.Sygkounas, E. Rizzo, G. Troncy, R.: Sentiment Polarity Detection from Amazon Reviews:
An Experimental Study. SemWebEval 2016. Comm. in Comp. and Info. Sc. 641, 108 – 120
(2016)
16.Shivaprasad, T.K., Shetty, J.: Sentiment analysis of product reviews: A review. ICICCT,
India (2017)
17.Leung, C. W. K., Chan, S. C. F.: Sentiment Analysis of Product Reviews. Encyclopedia of
Data Warehousing and Mining. 1794 – 1799 (2009)
18.Ray, P. Chakrabarti, A.: Twitter sentiment analysis for product review using lexicon
method. ICDMAI, India (2017)
19.Govindaraj, S. Gopalakrishnan, K.: Intensified Sentiment Analysis of Customer Product
Reviews Using Acoustic and Textual Features. ETRI J. 38, 494 – 501 (2016)
20.Rajkumar, V.: Sentiment analysis for Yelp review classification. Medium (2017)
21.McAuley, J., Yang, A.: Addressing Complex and Subjective Product-Related Queries with
Customer Reviews. WWW, Canada (2016)
22.Babu, U. R.: Sentiment Analysis of Reviews for E-Shopping Websites. IJECS. 6, 19965 -
19968 (2017)
23.Esuli, A., Sebastiani, F.: SENTIWORDNET: A Publicly Available Lexical Resource for
Opinion Mining. In Proc. of the 5th conf. on Lang. Resources and Eval. 1 - 6 (2016)
24.Baccianella, S., Esuli, A., Sebastiani, F: SENTIWORDNET 3.0: An Enhanced Lexical
Resource for Sentiment Analysis and Opinion Mining. In Proc. of the 5th conf. on Lang.
Resources and Eval. 1 – 5 (2010)
25.Kaur, B., Kumari, N.: Sentiment Analysis of Technical Article Review using Hybrid
Approach. IJARSSE 2016
26.Neo4J.: Sentiment Analysis for Amazon Reviews using Neo4J. Kvangundy (2015)
27.Li, L.: Identification of informative reviews enhanced by dependency parsing and sentiment
analysis. ICCCI, China (2016)
28.Bose, R., Dey, R. K., Roy, S., Sarddar, D.: Analyzing Political Sentiment using Twitter
Data. ICTIS, India (2018)
29.McAuley, J., Leskovec, J.: From amateurs to connoisseurs: modeling the evolution of user
expertise through online reviews. WWW, Canada (2013)
30.SNAP.: Amazon Fine Food Reviews. https://www.kaggle.com/snap/amazon-fine-food-
reviews Kaggle (2018)
31.Geetha, M., Singha, P., Sinha, S.: Relationship between customer sentiment and online
customer ratings for hotels – An empirical analysis. Tourism Mgmt. 61, 43 – 54 (2017)
32.Krouska, A. Troussas, C., Virvou, M.: The effect of preprocessing techniques on Twitter
sentiment. IISA, India (2016)
33.Zin, M. H., Mustapha, N., Murad, M. A. A., Sharef, N. M.: The effects of pre-processing
strategies in sentiment analysis of online movie reviews. AIP Conf. Proc. 1891, 020089
(2017)
34.Hemalatha, I., Varma, G. P. S., Govardhan, A.: Preprocessing the Informal Text for efficient
Sentiment Analysis. IJETTCS. 1, 58 – 61 (2012)
35.Word list. http://neuro.imm.dtu.dk/wiki/Word_list (2018)
36.Mohammad, S., Turney, P.: Emotions Evoked by Common Words and Phrases: Using
Mechanical Turk to Create an Emotion Lexicon. NAACL-HLT, LA, California. 1 - 9 (2010)
37.Nguyen, T. T., Chang, K., Hui, S. C.: Word Cloud Model for Text Categorization. ICDM,
Canada (2011)
38.Vithiya Ruba, K., Venkatesan, D.: Building a Customer Sentiment Analysis Tool based on
an Ontology for Twitter Posts. IJST. 8, 1 – 9 (2015)
39.Rajput, Q., Haider, S., Ghani, S.: Lexicon-Based Sentiment Analysis of Teachers’
Evaluation. Applied Computational Intelligence and Soft Computing, 2385429, 1 – 12
(2016)
40.Rambocas, M., Gama, J.: Marketing Research: The Role of Sentiment Analysis. FEP
Working Paper, n 489, 1 – 24 (2013)
41.Rambocas, M., Pacheco, B. G.: Online sentiment analysis in marketing research: a review. J.
of Research in Interactive Marketing. https://doi.org/10.1108/JRIM-05-2017-0030 (2018)
42.Songpan, W.: The Analysis and Prediction of Customer Review Rating Using Opinion
Mining. SERA 2017, London, UK, 71 – 77 (2017)
43.Li, Z., Liu, L., Li, C.: Analysis of customer satisfaction from chinese reviews using opinion
mining. Proc. of the 6th IEEE Int. Conf. on Software Engg. and Service Science (ICSESS),
95 – 99 (2015)
44.Wan, Y., Nakayama, M.: The Reliability Online Review Helpulness. J. of Electronic
Commerce Research, 15 179 – 169 (2014)
45.Amazon.in.: Amazon’s Top Customer Reviewers. https://www.amazon.in/review/top-
reviewers (2018)
46.Zhu, F, Liu, Q.: Competing with Complementors: An Empirical Look at Amazon.com.
Working Paper 15-044 1 – 34 (2016)
47.Nguyen, T. H., Shirai, K.: Topic Modeling based Sentiment Analysis on Social Media for
Stock Market Prediction. Proc. of the 53rd Annual Meeting of the Association for
Computational Linguistics and the 7th International Joint Conference on Natural Language
Processing, Beijing, China 1354 – 1364 (2015)
48.Bandhakavi, A., Wiratunga, N., Missie, S.: Lexicon Generation for Emotion Detection from
Text. IEEE Intelligent Sys, IEEE 102 – 108 (2017)