Multi-View Knowledge Graph Recommendation with User Review-Driven Preference Modeling

User-generated reviews, as interpretable behavioral proxies, encapsulate fine-grained preference patterns critical for enhancing recommendation accuracy in commercial systems. While traditional recommendation models face persistent challenges including data sparsity, cold-start dilemmas, and content homogenization, existing solutions predominantly rely on coarse historical interest modeling—an approach fundamentally limited in capturing dynamic intent evolution, thereby constraining personalization efficacy. In contrast, knowledge graph embedding techniques project heterogeneous entities and relations into unified low-dimensional spaces, achieving not only graph completion but also explicit modeling of cross-domain semantic interdependencies, thus providing transferable knowledge for recommendation tasks. To address the dual limitations of preference modeling inaccuracy and behavioral intent under-exploitation, we propose KGMR (Knowledge Graph-enhanced Multi-view Recommender), a hybrid architecture integrating multi-perspective relational embeddings with attention-based temporal dynamics modeling.