Transformer-Based Sentiment Analysis: A Comparative Study of Machine Learning, Deep Learning, and BERT Models on Amazon Product Reviews

Abstract

E-commerce platforms generate massive amounts of user-generated reviews, making sentiment analysis a crucial tool for understanding customer feedback. This study evaluates the effectiveness of traditional machine learning, deep learning, and Transformer-based approaches for sentiment classification of Amazon product reviews. Models including Multinomial Naive Bayes (MNB), Random Forest (RF), Long Short-Term Memory (LSTM), Convolutional Neural Networks (CNN), and fine-tuned BERT were analyzed. Reviews were preprocessed, and sentiment labels were derived from star ratings. Metrics such as Accuracy, Precision, Recall, F1-Score, and Area Under the Curve (AUC) were calculated to compare performance. Results revealed that the fine-tuned BERT model outperformed all others with an Accuracy of 97.4%, Precision of 96.8%, Recall of 98.2%, F1-Score of 97.5%, and AUC of 98.5%. Traditional ML models like MNB and RF lagged behind, achieving accuracies of 82.3% and 88.6%, respectively. This study highlights the superiority of Transformer-based architectures for sentiment analysis tasks and provides insights for future research in leveraging advanced NLP techniques.