.png){kind=icon}
.png){kind=icon}
Return
Multi-view hypergraph contrastive learning for bundle recommendation
.png){kind=icon}
PRE
AI
DOI：10.1007/s11227-025-07835-1.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Bundle recommendation aims to suggest coherent item sets to users, yet traditional approaches often struggle to model multi-entity interactions and inter-item synergies within bundles, leading to suboptimal performance. This paper introduces the multi-view hypergraph contrastive learning (MHCL) model to tackle these challenges. MHCL constructs three-view hypergraphs to explicitly capture user–bundle, user–item, and bundle–item relationships, effectively encoding high-order dependencies among entities. To aggregate information from these diverse views, the model incorporates an attention-based feature fusion mechanism, allowing it to learn nuanced user preferences. Additionally, MHCL utilizes contrastive learning to regularize the embedding space, thereby enhancing generalization and robustness, particularly in data-sparse scenarios. Experimental evaluations on three public datasets demonstrate MHCL’s superiority over baseline methods, achieving up to a 5.77% improvement in Recall@K and a 4.53% improvement in NDCG@K compared to the best performing baselines. These results underscore the efficacy of leveraging multi-view hypergraphs and contrastive learning to enhance bundle recommendation accuracy in scenarios characterized by complex entity relationships.
Keywords:
Recommender systems
Bundle recommendation
Hypergraph
Contrastive learning
Journal
T
IF:
0
Papers:
647
Citations:
0