Classification of product review sentiment by NLP and hybrid ML algorithms

Abstract

With the advent of e-commerce in Bangladesh, consumer sentiment can be better gauged through an uptick in reviews. Analyzing these evaluations is a challenge due to their disorganized nature. This study aims to improve sentiment classification in user-generated product evaluations by creating a hybrid machine-learning approach. The database has 5,000 reviews from various online sources, including Facebook groups and marketplaces like Daraz and Rokomari. Tokenization and other text preprocessing methods were employed. We established and evaluated a DT RF and SVM, a DT hybrid with NN and SVM, and an RF hybrid with SVM. I assessed these models using recall, accuracy, precision, and F1-score. With an F1 score of 0.95 and an accuracy of 94%, the hybrid DT, NN, and SVM models outperformed the other models. The DT, RF, and SVM models each received a score of 91% accuracy, while the RF and SVM models received a score of 97%. Based on the findings of this study, RF and SVM has the potential to enhance sentiment categorization for online purchases significantly. This, in turn, aids businesses in better understanding customer feedback and making informed decisions..