Abstract

In recent times, e-commerce has grown expeditiously. As a result, online shopping and online product reviews are increasing, which makes it nearly impossible for companies to analyze them. In addition, ratings with high star ratings are often ignored, which may contain dissatisfied reviews that should be taken into account. Therefore, techniques are required for companies to extract information from the reviews and ratings, which helps them to analyze the data and make accurate decisions. The objective of this paper is to compare supervised Machine Learning (ML) classification approaches on Amazon product reviews to determine which method offers the most reliable sentiment analysis results. The product reviews are pre-processed and the extracted sentiments are labelled as either positive or negative sentiments. The sentiments are analysed using Multinomial Naive Bayes (MNB), Random Forest (RF), Long-Short Term Memory (LSTM) and Convolutional Neural Network (CNN). The feature extraction techniques Term Frequency-Inverse Document Frequency Transformer (TF-IDF(T)) and TF-IDF Vectorizer (TF-IDF(V)) were used for ML models, MNB and RF. The performance of the models was evaluated using confusion matrix, Receiver Operating Characteristic (ROC), and Area under the Curve (AUC). The LSTM provided an accuracy of 97% and outperformed other models.