MUSE-Rec: Explainable multi-behavioral social e-commerce recommendation with integrated link prediction

Social e-commerce platforms create complex ecosystems where user behaviors, social relationships, and temporal dynamics are closely interconnected and evolve continuously over time. Current recommendation approaches face critical limitations: they struggle to model diverse user behaviors simultaneously, fail to predict evolving social connections, and lack interpretable explanations. Unlike existing methods that treat multi-behavioral modeling, social influence, and temporal dynamics as separate optimization problems, this work introduces MUSE-Rec, a unified Multi-behavioral Social E-commerce Recommendation Framework. MUSE-Rec integrates these interconnected components through an innovative graph neural network architecture. Integrating link prediction is crucial because predicting future social connections enables the system to anticipate how user preferences will evolve, improving recommendation accuracy and timing. Our framework advances recommendation systems theory by demonstrating that joint optimization of behavioral patterns, social dynamics, and temporal evolution achieves superior performance compared to component-wise approaches. This establishes new theoretical foundations for integrated social-temporal-behavioral modeling. MUSE-Rec introduces three key innovations: (1) a Multi-Graph Attention Network layer modeling diverse user-item interactions while predicting future social connections, achieving behavior correlation coefficient of 0.73 and link prediction AUC of 0.892; (2) an adaptive social connection aggregation mechanism capturing dynamic social influence patterns; and (3) a temporal graph network layer incorporating behavior-specific temporal dynamics. Comprehensive experiments on Yelp and Amazon Electronics datasets demonstrate superior performance. MUSE-Rec achieves NDCG@10 of 0.768 on Yelp and 0.742 on Amazon. The explainability module achieves high fidelity scores of 0.823 and 0.805 respectively, providing transparent behavior-specific explanations. MUSE-Rec enables e-commerce platforms to deploy more effective recommendation systems with 28% computational efficiency improvement while enhancing user trust.