Exploring Multiple Hypergraphs for Heterogeneous Graph Neural Networks

Ying Wang, Yingji Li, Yue Wu, Xin Wang

Abstract: Highlights•The hypergraph can preserve high-order proximity and capture semantic interactions.•Our model integrates multiple motif-based hypergraphs to cover all nodes in HIN.•Attention mechanism aggregates node features based on importance and semantic roles.•MoH-HGNN leverages hypergraph and attention to capture complex connectivity pattern.•Our model can be applied to multiple applications of heterogeneous networks.