Amazon Product Reviews: Sentiment Analysis Using Supervised Learning Algorithms

Abstract

E-commerce is gaining attraction in today's digital environment by making products closer to customers without forcing them to leave their homes. A customer has to go through hundreds of reviews before making a purchase. The number of reviews for a single product can easily approach millions and becomes difficult to understand the overall client feedback. However, machine learning techniques can understand and learn from big data. Consequently, sentiment analysis is a new study area combining natural language processing and text analytic to extract information from sources and classify the polarity of express sentiments. We have employed Support Vector Machine, Naive Bays, Decision Tree, Random Forest, Logistic Regression, and Multi-Layer Perceptron Classifiers for large-scale supervised learning on the Amazon Electronic products reviews data set and obtained satisfactory results. We have compared three different preprocessing techniques: TF-IDF, Bag of Words, and Word2Vec for representing the words in the review. MLP classifier produced the best results with Bag of Words preprocessing showing 92% accuracy.