Co-Representation Neural Hypergraph Diffusion for Edge-Dependent Node Classification

CoNHD is a novel hypergraph neural network approach for edge-dependent node classification that addresses limitations of existing methods by incorporating node-hyperedge co-representations and equivariant neural diffusion operators. It models within-edge and within-node interactions adaptively, allowing for edge-dependent messages and direct interactions. The model outperforms competitors on six real-world datasets, especially in scenarios with large-degree nodes and hyperedges, and demonstrates efficiency with UNP and ISAB operators. CoNHD's design, which extends hypergraph diffusion and employs set-equivariant networks, contributes to a more expressive and effective HGNN architecture for node classification tasks.