Sentiment Analysis for Amazon Product Reviews

Abstract

Sentiment analysis is a classicfication process whereby machine learning techniques are applied on text-driven datasets in order to analyse the emotion / opinion expressed in a text, e.g. a message being positive or negative about a certain topic. The problem is to conduct a sentiment analysis (positive and negative sentiment) on online product reviews of Products (unlocked mobile phones) sold on Amazon.com. The trained model can be used to predict users’ sentiment based on their online reviews. In this project, different machine learning algorithms are compared, trained and tested on a dataset containing 400000 reviews. The performance of three different algorithms were compared: Multinomial Naive Bayes (MNB), Logistic Regression and Long short-term memory network (LSTM). The Logistic Regression model resulted in the highest performance with Accuracy of 0.95 and AUC of 0.94. The dataset consists of 400 thousand reviews of products (unlocked mobile phones) sold on Amazon.com which is publicly available on Kaggle. Solution to the problem would be useful for a brand to gain a broad sense of user’s’ sentiment towards a product through online reviews Further study is needed to investigate if the classfication remains accurate when including more than two classes (e.g. Introducing a neutral class).