A Hybrid Deep Learning Approach for Recommendation Systems: Fusing Review Sentiment Analysis

Traditional approaches heavily rely on user ratings as the main source for generating recommendations. However, the inherent sparsity of ratings poses a significant challenge in delivering accurate and personalized recommendations. In this study, we propose an innovative and deep fusion personalized recommendation model that harnesses emotional features extracted from user reviews, with the goal of providing highly accurate and personalized recommendations. Our model incorporates cutting-edge techniques, including BERT and Bidirectional Recurrent Neural Network, which are renowned for their contextual understanding capabilities. By leveraging these advanced techniques, we can capture the fine-grained sentiment information embedded within user reviews, enabling comprehensive understanding of the conveyed emotional aspects. Furthermore, we introduce an enhanced feature weight calculation method that synergistically integrates lexicon-based techniques and attention mechanisms. Through meticulous experimentation on a comprehensive and diverse dataset, we empirically demonstrate the efficacy of our proposed deep fusion model. The experimental results validate its superior performance compared to various baseline models, showcasing substantial improvements. By leveraging emotional features extracted from user reviews, our sophisticated model empowers e-commerce platforms to offer highly reliable and personalized recommendation services.