ABSTRACT Recommendation systems are crucial for filtering information, but traditional methods like collaborative filtering (CF) often struggle with data sparsity and understanding nuanced user interests based solely on ratings or surface‐level review analysis. To address these limitations, we propose an innovative hybrid recommendation model (APBHM). This model integrates multi‐level user and item information by employing a weighted combination strategy that fuses two key components: (1) Fine‐grained attribute preferences derived from analyzing both user viewing history (using Latent Dirichlet Allocation—LDA) and user‐generated reviews (using Named Entity Recognition, Sentiment Analysis, and LDA to capture aspect opinions); and (2) Preference scores generated by traditional user‐based collaborative filtering. By learning user preferences for specific item attributes (e.g., actors, directors) from multiple sources, APBHM captures a deeper understanding of user interests. Experimental results on a real‐world movie dataset demonstrate the effectiveness of our approach; APBHM significantly outperforms the traditional user‐based CF baseline, achieving relative improvements of approximately 26% in Precision and 31% in Recall for Top‐3 recommendations. The findings confirm that our hybrid model successfully combines the strengths of content‐based and collaborative filtering approaches, effectively mitigating data sparsity issues and yielding higher quality, more personalized recommendations by accurately identifying user preferences for fine‐grained item attributes.
Yang et al. (Sun,) studied this question.