This study presents a hybrid sentiment analysis framework combining rule-based and transformer-based natural language processing techniques for large-scale e-commerce review classification. The research integrates VADER for lexicon-driven sentiment scoring and RoBERTa for contextual deep learning-based classification to improve prediction accuracy and robustness. The model was trained and evaluated on a dataset of 90,000 customer reviews, enabling reliable performance assessment across multiple evaluation metrics. Experimental results demonstrate strong classification effectiveness, achieving an accuracy of 88.4%, precision of 89.2%, recall of 88.1%, and an F1-score of 88.6%. In addition to predictive modeling, this research incorporates data visualization techniques using Power BI to provide interpretable business insights derived from sentiment trends. The proposed framework highlights the practical application of modern NLP techniques in marketing analytics, customer feedback interpretation, and decision-support systems. This work contributes to applied machine learning research by demonstrating the effectiveness of hybrid sentiment analysis approaches that combine lexicon-based heuristics with transformer-based models.
Aishwarya Patil (Wed,) studied this question.