Position-Awareness and Hypergraph Contrastive Learning for Multi-Behavior Sequence Recommendation
Sitong Yan, Chao Zhao, Ningning Shen, Shaopeng Jiang · IEEE Access · 2024
Existing multi-behavior sequential recommendation methods obtain users’ interest preferences by analyzing their historical multi-behavior information to uncover users’ potential intentions in multi-behavior sequential recommendation. However, the existing methods still have problems such as unstable users’ interest preferences and difficulty in capturing the fine-grained relationships between behaviors. This paper proposes a framework called Position-aware Hypergraph Contrastive Learning for Multi-Behavior Sequence Recommendation (PHCL-MBSR). It introduces position context encoding in the temporal dimension, focusing on the context behavior dependencies. In the spatial dimension, a global hypergraph is constructed to capture the high-order relationships between sequences, and the global multi-behavior dependencies in the spatial dimension are captured through hypergraph contrastive learning. PHCL-MBSR has been experimentally evaluated on three benchmark datasets, and the results demonstrate the effectiveness and interpretability of this framework.