A Multi-View Feature Fusion-Based Recommendation Model with User-Side Data Mining

Traditional recommender systems typically rely on collaborative filtering (CF) methods based on user-item interactions. However, these methods often face data sparsity issues, which negatively impact recommendation performance and degrade user experience. Recent research has focused on methods that incorporate side information to enhance representation learning and improve data utilization. To this end, this paper proposes a recommendation model named RMSI, aimed at mining and fusing user-side information. To construct this model, we utilize multi-faceted information to build user-item interaction graphs (including view, click, add to cart, etc.), user-user social network graphs, and item-category information graphs. An encoder is used to capture different behaviour relationship between users and items, and a graph neural network is used to capture the strength of user-user social relationship. Finally, an attention mechanism is used to weight and integrate the user representations extracted from these different graphs. Experimental results on synthetic and real-world datasets demonstrate that RMSI achieves better recommendation performance.