Research map / Graph learning / GNN architectures
Enhancing Graph Representation Learning with Localized Topological Features
Zuoyu Yan, Qi Zhao, Ze Ye, Tengfei Ma, Liangcai Gao, Zhi Tang, Yusu Wang, Chao Chen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| add pair-wise topological feature instead of node-wise concatenation (PEGN vs PEGN (node-wise)) (ablation) | PEGN · Computers link prediction | AUC-ROC: 97.0 -> 97.9 | Improved |
| add pair-wise topological feature instead of node classification features (PEGN vs PEGN (NC)) (ablation) | PEGN · Cora link prediction | AUC-ROC: 93.0 -> 94.9 | Improved |
| add pair-wise topological feature instead of node-wise concatenation (ablation) | PEGN · Citeseer link prediction | AUC-ROC: 92.9 -> 95.1 | Improved |
| use learnable filter function instead of Ollivier-Ricci curvature (PEGN (FF)) | PEGN · Cora node classification | accuracy: 82.7 -> 81.8 | Hurt |
| use learnable filter function instead of Ollivier-Ricci curvature (PEGN (FF)) | PEGN · Computers node classification | accuracy: 86.6 -> 86.2 | Hurt |
| combine persistence images (PEGN (PI)) instead of curvature filter | PEGN · Photo node classification | accuracy: 92.7 -> 90.2 | Hurt |
| use persistence image combination (PEGN (PI)) in link prediction | PEGN · Computers link prediction | AUC-ROC: 97.9 -> 98.3 | Improved |
| use learnable filter function (PEGN (FF)) in link prediction | PEGN · PPI link prediction (graph 1) | AUC-ROC: 83.71 -> 85.47 | Improved |
| use PI+ representation instead of original persistence image with GCN (ablation) | GCN · Photo node classification | accuracy: 92.5 -> 93.0 | Improved |
| use PI+ representation instead of original persistence image with GCN (ablation) | GCN · CS node classification | accuracy: 92.3 -> 93.0 | Improved |
| add persistence-image topological features to GCN (GCN(+PI+) vs GCN) (ablation) | GCN · Computers node classification | accuracy: 82.6 -> 86.9 | Improved |
| swap in the relaxed shortest-path filter used in the proof (PEGN (new)) (ablation) | PEGN · Physics node classification | accuracy: 94.3 -> 94.4 | No clear effect |
Closest papers by meaning
- Graph Persistence goes Spectral (NeurIPS 2025 · similarity 0.89)
- TopoGCL: Topological Graph Contrastive Learning (AAAI 2024 · similarity 0.88)
- Weisfeiler and Lehman Go Paths: Learning Topological Features via Path Complexes (AAAI 2024 · similarity 0.88)
- Graphcode: Learning from multiparameter persistent homology using graph neural networks (NeurIPS 2024 · similarity 0.88)
- TopInG: Topologically Interpretable Graph Learning via Persistent Rationale Filtration (ICML 2025 · similarity 0.88)
- TopER: Topological Embeddings in Graph Representation Learning (NeurIPS 2025 · similarity 0.88)
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.
