Knowledge Recommendation Model with Semantic Alignment Based on Hypergraph Convolutional Networks

Knowledge graphs (KGs) have emerged as a mainstream solution to alleviate data sparsity and cold-start problems in recommender systems. However, existing KG-based recommendation methods face challenges in capturing the complex collaborative relations within user-item interaction graphs and mitigating heterogeneous semantic biases between interaction and KGs. To address these limitations, this paper proposes a Knowledge Recommendation Model with Semantic Alignment Based on Hypergraph Convolutional Networks (SAB-HCN). Specifically, this paper constructs local graphs from useritem interactions and corresponding KG facts, and non-local graphs from co-occurrence relationships and additional knowledge in the user-item-entity graph. To jointly model higher-order interest associations and latent behavioral features, this paper builds global and non-global hypergraphs to capture both global interactions and local neighborhood signals. User and item embeddings are then learned through a hierarchical hypergraph convolution network. To bridge semantic gaps between different graphs, a cross-graph semantic alignment mechanism is introduced, using a learnable mapping function to align item embeddings across heterogeneous spaces. Experimental results show that SAB-HCN consistently outperforms state-of-the-art baselines on both click-through rate prediction and Top-K recommendation. On the MovieLens-1M dataset, SAB-HCN achieves AUC and F1 improvements of 1.43% and 1.77%, respectively.