International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169

Volume: 5 Issue: 7 28 31
_______________________________________________________________________________________________

Polarity Categorization on Product Reviews

Mixymol V.K.
Assistant Professor,
Department of Computer Science,
Kanchi Mamunivar Centre for Post Graduate Studies,
Puducherry-605008,
India. Email : [email protected]
Mobile : 9443430086

Abstract: People search for other peoples opinions from the internet before purchasing a product, when they are not familiar about a specific
product. With the help of reviews, ratings etc. online data presents useful information to customers for buying a product and for manufacturers to
improve the quality of product. When an individual wants to make a decision about buying a product or using a service, they have access to a
huge number of user reviews, but reading and analyzing all of them is a tedious task. Reading all of them is generally inefficient. There is a need
for summarization in product reviews. Sentimental analysis helps customer visualize satisfaction while purchasing by simple summarization of
these reviews into positive or negative two broader classified classes. The study aims to tackle the problem of sentiment polarity categorization.
The data set is collected from amazon.com. The data set contains 376 instances of reviews of Nokia mobile in the form of a text file. Two
classification algorithms namely Nave Bayes and Support Vector Machine Algorithms are taken to classify the reviews as positive, negative or
neutral.

Keywords: Product review, Sentiment analysis, Classification, Polarity Categorization.

__________________________________________________*****_________________________________________________

I. INTRODUCTION satisfaction while purchasing by simple summarization of

In past days, buying of products was more based on these reviews into positive or negative two broader classified
getting product review from nearby neighbors, relatives etc. as classes. Comments are mainly used for helping customers
products were purchased directly from merchants. But with purchase online and for knowing current market trends about
change in technology, development of E-commerce industry products which is helpful for developing market strategies by
with sites flooded by products from different brands made merchants.
available to customers at the touch of one click.[1] The There are several reasons why potential buyers read
availability of product based sites with doorway delivery has product reviews, Common questions readers would like to
made it convenient for customers to shop online. With so know include:
much change in shopping pattern, merchants providing Is the product easy to use?
customers with feedback option about the product. Customers Is it of a high quality?
write reviews from all parts of the world. A lot of reviews are Is it geared towards somebody like me?
very long, making it hard for a potential customer to review Have others had a good experience with the product?
them to make an informed decision on whether the customer What are the pros and cons of the product?
should purchase the product or not. A huge number of reviews What alternatives are available, and how do they measure
also make it difficult for product manufacturers to keep log of up?
customer opinions and sentiments expressed on their products
Is the product worth my money?
and services. It thus becomes necessity to produce a summary
of reviews. Summarization of reviews is done using sentiment
II. PROBLEM STATEMENT
analysis.[2]
One of the most common applications of sentiment
Sentiment analysis tends to extract subjective information
analysis is the area of reviews of consumer products and
required for source materials by applying natural concept of
services. There are many websites that provide automated
natural language processing. The main task lies in identifying
summaries of reviews about products and about their specific
whether the opinion stated is excellent, good or bad. Since aspects. This work focuses on analyzing product based
customers usually do not express opinions in simple manner,
reviews by classifying them as positive, negative and neutral.
sometimes it becomes tedious task to judge an opinion stated.
Dataset for the study is collected from Amazon.com and the
Some opinions are comparative ones while others are
dataset contains 376 reviews of Nokia mobile. The dataset
direct.[3] Sentimental analysis helps customer visualize
28
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 7 28 31
_______________________________________________________________________________________________
contains unstructured data which is in the form of text file. NAs and zero length characters appropriately. Stringr also
Implementation of the work has done in R programming ensures that the data output from each function matches the
language. Various classification rules namely, Nave Bayes input data structures of other functions.
Classifier and Support Vector Machine are used to train a ggplot2 This package provides an implementation of the
sample dataset and measure the Accuracy value using R grammar of graphics in R, combining the advantages of
programming language. both base and lattice graphics. Plots can be built up step by
step from multiple data sources.
III. DATA DESCRIPTION e1071 this package provides function for latent class
The dataset for the project work is collected from analysis, such as Support vector Machines, bagged
Amazon.com. The data is collected from clustering and Naive Bayes classification.
https://github.com/SamTube405/Amazon-E-commerce-Data- Figure 1 gives the step wise activities of the work done.
set. A subset of the dataset which consists of 376 reviews for
Nokia mobile is considered for this study. Reviews are in the
form of a text file and the reviews include product, user
information, rating, plain text, and it contains special
characters symbols.
Sample product reviews:
best 4mp compact digital available
camera is perfect for an enthusiastic amateur
photographer
macro[+3]##the pictures are razor-sharp , even in
macro
it is small enough to fit easily in a coat pocket or
purse
it is light enough to carry around all day without
bother

Experimentation and Result analysis

R programming Language is used to implement the Pre-processing
classification algorithms. R is a programming language for The dataset is unstructured; it may contain repetitive
statistical calculation and it is created by statistician for make words, large number of words that are not at all needed in
statistical data analysis easier. Rstudio is an Integrated summarizing of opinions. Pre-processing involves removal of
Development Environment (IDE) that helps to develop stop words such as and, or, that etc. followed by porter
programs in R. And it is free and open source software. And stemming which involves simplifying target words to base
also it is used for graphics representation and reporting. words by removal of suffixes such as ed, ate, ion, ional,
Packages used for this study includes ment, ator, ssess, es, ance or conversion from ator to ate etc.
NLP - Natural Language Processing For example, replacement is stemmed to replac; troubled
tm - text mining packages to trouble ; happy to happi ; operator to operate. The raw
RTextTools-It is a machine learning package for data is pre-processed to improve quality.
automatic text classification
RColorBrewer-It is an R packages that uses the work Feature Extraction
from to help you choose sensible color schemes for figures Features in reviews are extracted so that it helps
in R customer to know which feature has positive comment and
Bitops- Doing Bitwise operations which one has negative. Since, overall conclusion about
plyr The package plyr is a set of tools that solves common product is much needed but there is also situation where
problems by breaking down bigger problems into more customer requirements come into the scenario. Use of
workable pieces. The package then operates on each adjectives is done to classify opinions as positive or negative
problem before reassembling the reworked pieces back using unigram model. For example, the Samsung camera I
together. bought was good; it has got great touch screen, awesome
Stringr stringr makes R string functions more consistent, flashlight. The feature extracted out of it would be like:
simpler and easier to use by ensuring the function and Domain: Mobile; Product: Samsung; Feature: Camera;
argument names are consistent and all functions deal with Adjective: Good.

29
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 7 28 31
_______________________________________________________________________________________________
Training and classification
Supervised learning generates a function which maps
inputs to desired outputs also called as labels because they are
training examples labelled by human experts. Here, nave
Bayes and Support Vector Machine techniques to carry out
supervised learning on the dataset fetched.

Support Vector Machine:

Support Vector Machines (SVM) is binary classifies
is able to classify data samples in two disjoint classes. The
basic idea behind is classifying the sample data into linearly
separable. Support Vector Machines (SVMs) are a set of
related supervised learning methods used for classification and
regression. In simple words given a set of training examples,
each marked as belonging to one of two categories. An SVM
training algorithm builds a model that predicts whether a new Figure 2 Classified as positive, negative and neutral
example falls into one category or the other. SVM is a set of
related supervised learning method for classification and
regression.[4] SVM simultaneously minimize the empirical
classification error and maximize the geometric margin. SVM
is called Maximum Margin Classifiers and it can be efficiently
perform non-linear classification using kernel trick.

Nave Bayes Classifier

The Naive Bayes algorithm is based on conditional
probabilities. It uses Bayes' Theorem, a formula that calculates
a probability by counting the frequency of values and
combinations of values in the historical data. Bayes' Theorem
finds the probability of an event occurring given the
probability of another event that has already occurred. The
Naive Bayesian classifier is based on Bayes theorem with
independence assumptions between predictors. [5]A Naive
Bayesian model is easy to build, with no complicated iterative Figure 3 Algorithm comparison
parameter estimation which makes it particularly useful for
very large datasets. Despite its simplicity, the Naive Bayesian IV. CONCLUSION and Future Directions
classifier often does surprisingly well and is widely used Sentiment analysis deals with identifying and
because it often outperforms more sophisticated classification aggregating the sentiment or opinions expressed by the users.
methods. Sentiment analysis is to classify the polarity of text in
Fgure 2 gives the classification of these input set into document or sentence whether the opinion expressed is
positive, negative and neutral and figure 3 shows the positive, negative, or neutral. Here two approaches have been
performances of these algorithms are measured by the compared and a result for both approaches on the product
accuracy value. The comparative study shows that SVM review dataset has been done. SVM is found to give better
performs better than Naive Bayes for Sentiment analysis on accuracy that is 100% as compared to Nave Bayes approach
product reviews. which is 80%. The study can be extended to create manual
dictionaries, like gd,( for good, xlnt ( for excellent ), and
using other sybols like !!!, * etc.,

REFERENCES
[1] Manvee Chauhan, Divakar Yadav Sentimental Analysis of
Product Based Reviews Using Machine Learning
Approaches Journal of Network Communications and
Emerging Technologies (JNCET) Volume 5, Special Issue 2,
December (2015) .

30
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________
International Journal on Recent and Innovation Trends in Computing and Communication ISSN: 2321-8169
Volume: 5 Issue: 7 28 31
_______________________________________________________________________________________________
[2] Pravesh Kumar Singh1, Mohd Shahid Husain2
Methodological study of opinion mining and sentiment
analysis techniques International Journal on Soft Computing
(IJSC) Vol. 5, No. 1, February 2014.
[3] Amit Gupte, Sourabh Joshi, Pratik Gadgul, Akshay Kadam
Comparative Study of Classification Algorithms used in
Sentiment Analysis Amit Gupte et al, / (IJCSIT)
International Journal of Computer Science and Information
Technologies, Vol. 5 (5) , 2014, 6261-6264.
[4] https://www.analyticsvidhya.com/blog/2015/10/understaing-
support-vector-machine-example-code
[5] http://software.ucv.ro/~cmihaescu/ro/teaching/AIR/docs/Lab4
-NaiveBayes.pdf.

31
IJRITCC | July 2017, Available @ http://www.ijritcc.org
_______________________________________________________________________________________