Distilled Bidirectional Encoder Representations from Transformers for Multi-Class Multi-Label Sentiment Analysis in Product Review

The rapid growth of e-commerce has revolutionized how consumers shop and share feedback, turning product reviews into a valuable source of authentic customer sentiment. This research focuses on analyzing these reviews for sentiment across multiple product aspects, specifically addressing the complex multi-class, multi-label sentiment analysis of laptop reviews. This study rigorously evaluated a range of transformer model's Bidirectional Encoder Representations from Transformers (BERT), Robustly Optimized BERT Pretraining Approach (RoBERTa), Distilled BERT (DistilBERT), and A Lite BERT (ALBERT) to determine the most effective model for capturing nuanced sentiment in multiple categories. The DistilBERT emerged as the top performer, achieving a notable accuracy of 95.27%, precision of 90.76%, recall of 95.27%, and Fl-score of 92.96%, highlighting its potential for detailed sentiment analysis. This study not only underscores the value of advanced transformers in understanding diverse sentiment nuances but also contributes insights into handling multi-class, multi-label challenges in sentiment classification.