Reproduce “Homophily-oriented Heterogeneous Graph Rewiring” 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
Jiayan Guo, Lun Du, Wendong Bi, Qiang Fu, Xiaojun Ma, Xu Chen, Shi Han, Dongmei Zhang and 1 more
Homophily-oriented Deep Heterogeneous Graph Rewiring (HDHGR) enhances the effective homophily of heterogeneous graphs, leading to improved performance for various existing Heterogeneous Graph Neural Networks (HGNNs). The method achieves consistent gains, especially on heterophilous datasets, by selectively adding edges between similar nodes and removing edges between dissimilar ones.
Key insights
- Existing Heterogeneous Graph Neural Networks (HGNNs) perform poorly on heterophilous heterogeneous graphs because they implicitly assume homophily, particularly along meta-paths.
- Graph rewiring, by intelligently adding and removing edges based on learned node similarities, can effectively enhance the homophily of meta-path subgraphs.
- A theoretical analysis indicates that increasing homophily along meta-paths can improve the generalization ability of HGNNs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.