Reproduce “Classification of Edge-dependent Labels of Nodes in Hypergraphs” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Minyoung Choe, Sunwoo Kim, Jaemin Yoo, Kijung Shin
Researchers from KAIST and Carnegie Mellon University introduce and tackle the problem of classifying edge-dependent labels for nodes in hypergraphs, where a node's role varies based on the hyperedge it belongs to. They propose WHATsNet, a hypergraph neural network that utilizes within-hyperedge attention and centrality-based positional encoding to achieve superior performance on this task and demonstrate utility in downstream applications like ranking and clustering.
Key insights
- Edge-dependent node labels represent a distinct and important characteristic of real-world hypergraphs that cannot be adequately captured by existing HGNNs focusing on global node properties.
- Explicitly modeling the relationships between nodes *within* each hyperedge (via `WithinATT`) is crucial for understanding and predicting a node's context-specific role.
- Incorporating relative positional information based on a node's centrality *within* a hyperedge (via `WithinOrderPE`) significantly enhances the learning of edge-dependent features.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.