Importance of Neutrality in Sentiment Analysis: A
Case Study of Amazon Food Reviews
School of Computing
National College of Ireland
Dublin
Abstract—In this era of social media reviews play a
major role specifically in B2C and C2C business models.
Reviews are given utmost importance as they affect the
business performance directly. Most studies in respect to
sentiment analysis ignore the neutral reviews thinking that
vote of majority has greater impact and that there isnt
much to learn from neutral texts. In this work, we propose
sentiment analysis considering neutral reviews rather the
rational way of analysis using Machine learning techniques
Random Forest, Nave Bayes and Radial SVM. We evalu-
ated their performance using k-fold cross validation and
found that Random Forest outperforms others. Finally,
we will see the importance, effects and advantages of
considering neutrality in business perspective.
Index Terms—Sentiment Analysis, Neutrality, Random
Forest , SVM, Naive Bayes , Hyper parameter Tuning
I. INTRODUCTION
Whats others thinking This is an important aspect for
many people in the world and plays a crucial role in de-
cision making. Each and everyday people discuss many
things and shed their views on various social platforms.
Many companies find different ways to gather this data
as it is rich in content and can help in growing their
business. Sentiment analysis is a method of mining and
obtaining information with regards to public perspective
about a particular topic. In order to reap full benefits of
this, these days many E-commerce sites are providing a
social platform integrated with their current services and
people are opting for this because of convenience,door
delivery etc. There is a lot of chances for the customers to
get tricked as they dont experience the product they are
going to buy. In this virtual world of shopping, reviews
play an important role in telling new buyers about their
experience with the product which they have already
bought and provided a rating for it. These reviews and
ratings are important for new buyers in finding how
worthy a product is? Hence analysing these reviews and
educating the buyers with sentiment of people helps them
in understanding about product better. On the other hand,
organisations can also learn how people are reacting
for their product in different markets. This information
provides exact facts and opinions of about the product
better than any sales or marketing analyses.
In general sentiment analysis is of three types as-
pect level, document level and sentence level [1]. In
document-level analysis it is considered to be of a
single topic and analysis is done. Similarly, sentences
are considered and analysed. Naturally when compared,
between these two there isnt much difference as sen-
tences are a part of document. In aspect level, it is at the
minimum grain level. In these types of analysis, there
are a lot of disadvantages and care has to be taken when
considering the data as reviews can be posted by anyone
so chances of fake reviews are more which directly affect
the sentiment analysis. The data that has been considered
in this project is from Amazon [2]. The reviews posted
are by its customers who have bought and experienced
the product.
In general, research on sentiment analysis is per-
formed as binary classification considering the positive
and negative reviews. They ignore the neutral reviews
leading inappropriate results. In our case study, we are
investigating the importance of neutrality and impact
of ignoring neutral reviews in sentiment classification
on accuracies of model.In this project, we are going
to see how sentiment analysis varies when neutrality is
considered and how it varies when neutrality is ignored.
Our objective is to say how the effect of neutrality on
model performance and depending the performance of
model, we will deliver meaningful insights to enhance
the performance of amazon food industry.
This report is organised as follows: Section II ex-
plains related work and model selection. Section III is
about methodology chosen. Section IV explains about