Sentiment Our Recommended System For E-Commerce Platform Using Large Language Model

Abstract

Background: With the rapid expansion of e-commerce platforms, personalized product recommendation systems have become essential for enhancing user satisfaction and business performance. Traditional systems primarily rely on user-item interactions, often neglecting the emotional context present in customer reviews.

Problem: Conventional recommendation models struggle to capture sentiment-driven user preferences, leading to less relevant or generic suggestions. Moreover, the lack of optimization and contextual understanding limits the adaptability and accuracy of these systems.

Methods: To address this gap, we propose a hybrid sentiment-aware recommendation framework combining Bidirectional Encoder Representations from Transformers (BERT) for deep sentiment feature extraction, the Marine Predators Algorithm (MPA) for optimization, and a Decision Tree (DT) and clustering for interpretable classification. The model integrates sentiment scores into the recommendation pipeline to refine user profiling and improve suggestion relevance.

Results: The proposed MPA_BERT+DT system was evaluated on datasets from three real-world platforms like Chase Technologies, SJ Enterprise, and Order Your Choice. It outperformed existing algorithms such as Naive Bayes, SVM, Random Forest, and baseline BERT, achieving an accuracy of 94.0%, with statistically significant improvements in precision, recall, and F1-score.

Conclusion: The MPA_BERT+DT model delivers a scalable and effective recommendation approach that combines sentiment analysis, optimization, and explainable classification. It enhances personalization, increases user trust, and demonstrates clear potential for real-world deployment in sentiment-rich e-commerce environments.