Sentiment Polarity Detection Using Machine Learning and Deep Learning

Abstract

As e-commerce has grown in recent years, so online shopping has increased with the number of product reviews posted online. The consumer's recommendations or complaints influence significantly customers and their decision to purchase. Sentiment polarity analysis is the interpretation and classification of text-based data. The main goal of our work is to categorize each customer's review into a class that represents its quality (positive or negative). Our sentiment polarity detection consists of the following steps: preprocessing, feature extraction, training, classification and generalization. First, the reviews were transformed into vector representation using different techniques of Tf-Idf and Tokenizer. Then, we trained with a machine learning model of SVM Linear, RBF, Sigmoid kernel and a deep learning model LSTM. After that, we evaluated the models using accuracy, f1-score, precision, recall. Our LSTM model predicts an accuracy of 86% for Amazon-based customer reviews and an accuracy of 85% for Yelp customer reviews.