Classifying the sentiment polarity of Amazon mobile phone reviews and their ratings

Abstract

After their purchase, reviews posted by customers are an important metric that businesses can use to their advantage for product improvement. On e-commerce websites, customers can give star ratings and text reviews. While star ratings sometimes simplify the analysis, the text reviews' detailed sentiment analysis can give deeper insights. This research aims to study the consumer sentiment on mobile phone reviews, their classification into the different star ratings (on a five-point scale), and determine the sentiment's accuracy with the star ratings. To achieve this, first, the data was pre-processed and cleaned. Then the text was translated into their numerical values using word embedding and the TF-IDF feature extraction method. Finally, different algorithms like SVM, Naϊve Bayes, Stochastic Gradient Descent, Logistic Regression, and Ensemble models were used, and performance accuracies were compared. The evaluation metrics used are accuracy, precision, recall, and the Fl-Score. It is found that a Random Forest classifier with Unigram on a balanced dataset gives better performance.