Research map / Graph learning / GNN architectures
A critical look at the evaluation of GNNs under heterophily: Are we really making progress?
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| remove duplicate nodes (filtered datasets) (ablation) | squirrel and chameleon, FSGNN · squirrel/chameleon | accuracy: squirrel 68.93 -> 35.92; chameleon 77.85 -> 40.61 | Hurt |
| remove duplicate nodes (ablation) | ResNet+adj on chameleon · chameleon | accuracy: 71.07 -> 38.67 | Hurt |
| separate ego- and neighbor-embeddings (GAT-sep vs GAT) | GAT on roman-empire · roman-empire | accuracy: 80.87 -> 88.75 | Helped |
| separate ego- and neighbor-embeddings (GT-sep vs GT) | Graph Transformer on roman-empire · roman-empire | accuracy: 86.51 -> 87.32 | Helped |
| separate ego- and neighbor-embeddings (GAT-sep vs GAT) | GAT on minesweeper · minesweeper | ROC AUC: 92.01 -> 93.91 | Helped |
| separate ego- and neighbor-embeddings (GAT-sep vs GAT) | GAT on amazon-ratings · amazon-ratings | accuracy: 49.09 -> 52.70 | Helped |
| separate ego- and neighbor-embeddings (GT-sep vs GT) | Graph Transformer on questions · questions | ROC AUC: 77.95 -> 78.05 | Helped |
| standard GNNs vs heterophily-specific models on new benchmark | baselines (GCN, SAGE, GAT, GT variants) vs specialized models · proposed 5 heterophilous datasets | accuracy/ROC AUC | Helped |
| graph-agnostic ResNet vs graph models on new datasets | roman-empire, minesweeper, questions · proposed datasets | accuracy/ROC AUC | No clear effect |
Closest papers by meaning
- On the Impact of Feature Heterophily on Link Prediction with Graph Neural Networks (NeurIPS 2024 · similarity 0.93)
- Understanding Heterophily for Graph Neural Networks (ICML 2024 · similarity 0.92)
- Classic GNNs are Strong Baselines: Reassessing GNNs for Node Classification (NeurIPS 2024 · similarity 0.91)
- What Is Missing For Graph Homophily? Disentangling Graph Homophily For Graph Neural Networks (NeurIPS 2024 · similarity 0.91)
- Making Classic GNNs Strong Baselines Across Varying Homophily: A Smoothness–Generalization Perspective (NeurIPS 2025 · similarity 0.91)
- Integrating Co-Training with Edge Discrimination to Enhance Graph Neural Networks Under Heterophily (AAAI 2025 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.