Relational Perception and Semantic Perception-Based Heterogeneous Graph Neural Network for Product Recommendation

To address the complexity of user-product interactions and semantic diversity in multi-relational heterogeneous graphs, this paper proposes a heterogeneous graph neural network recommendation framework (RSA-HGNN4Rec) that integrates relational perception and semantic perception. The model realizes fine-grained relational interaction modeling and dynamic high-order semantic path learning through a dual-path mechanism, and introduces a gating fusion mechanism to adaptively integrate multi-source information. To further validate the model efficiency, this paper compares the computational complexity of RSA-HGNN4Rec with mainstream models and provides a visualization analysis of the semantic path attention weights. Experiments on multiple real datasets, including Amazon-MR, ACM-MR, and DBLP-MR, demonstrate that RSA-HGNN4Rec significantly outperforms mainstream baseline models in terms of Recall@20 and NDCG@20 metrics for product recommendation tasks, while also demonstrating excellent generalization ability and interpretability on node classification tasks.