Hypergraph learning model offers clear insight into complex network data
Interpretable Hypergraph Learning via Neural Additive Models
Machine LearningSocial and Information Networks
Summary
Networks often have complex relationships involving groups rather than just pairs. Existing methods predict well but act like black boxes, making it hard to understand why. The authors developed a new model, HGNAN, that breaks down features and relationships in a clear way while keeping accuracy high. This helps explain predictions about nodes and connections in networks with complex group ties.
What this means in practice
- •For data scientists: Use HGNAN to build predictive models on complex network data with clear explanations for feature and structural influence.
- •For social network analysts: Analyze group interactions within social networks while understanding the impact of different factors on predictions.
Authors
Shihan Feng, Xin Zheng, Shiyi Yang, Ren Wang, Chudi Zhong, Can Chen
Abstract
Hypergraphs offer a natural framework for modeling networked data, where dependencies among entities are governed by higher-order interactions. While hypergraph learning methods such as hypergraph neural networks have demonstrated remarkable predictive performance, most existing approaches rely on black-box message-passing architectures, making it difficult to disentangle the contributions of node attributes and higher-order structural information. To address this challenge, we introduce the hypergraph neural additive network (HGNAN), an inherently interpretable framework for learning on hypergraph-structured data. HGNAN extends classical neural additive models to higher-order relational data by integrating feature-wise nonlinear decomposition with hypergraph-aware structural aggregation, enabling transparent prediction for both node- and hyperedge-level tasks. Extensive experiments on benchmark datasets demonstrate that HGNAN achieves performance comparable with state-of-the-art hypergraph learning methods while providing intrinsic and meaningful interpretability.