层次拓扑增强的多模态推荐方法OA

Recommendation systems often suffer from insufficient utilization of multi-modal information and noisy interaction data.To address these issues,this paper proposes a hierarchical topology-enhanced multi-modal recommendation method from an interac-tion denoising perspective.The method employs hierarchical anchor topology embedding,which performs stratified anchor sam-pling based on node activity levels to model global relationships,thereby mitigating over-smoothing and capturing long-range depen-dencies.Based on self-supervised learning,a learnable generator is designed to produce user multi-modal preferences and denoised interaction graphs,enabling bidirectional collaborative updating between user preferences and interaction graphs.A"graph-structure-modality"consistency constraint is incorporated to effectively suppress interaction noise.Experiments conducted on three real-world datasets demonstrate that the proposed method achieves average improvements of 2.37%and 3.27%on Recall@20 and NDCG@20 metrics,respectively,compared with 13 baseline models.