Research map / Graph learning / GNN architectures
Exploiting Edge Features for Graph Neural Networks
Liyu Gong, Qiang Cheng
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 |
|---|---|---|---|
| Use the edge-feature enhanced EGNN(C) instead of GCN | GCN · Cora dense | accuracy: 87.6\pm{}0.6\% -> 88.8\pm{}0.3\% | Improved |
| Use EGNN(C) with multidimensional molecular bond features | Weave · Tox21 test | AUC: 0.80\pm{}0.02 -> \bm{0.82}\pm{}0.01 | Improved |
| Use EGNN(C) with multidimensional molecular bond features | RF · Lipo test | RMSE: 0.86\pm{}0.04 -> \bm{0.75}\pm{}0.01 | Improved |
| Use EGNN(A) with multidimensional molecular bond features | Weave · Freesolv test | RMSE: 1.37\pm{}0.14 -> \bm{1.01}\pm{}0.12 | Improved |
| Add doubly stochastic normalization (ablation) | EGNN(C) · Cora dense | accuracy: 86.1\pm{}0.5\% -> 88.8\pm{}0.3\% | Improved |
| Add multidimensional directed-edge features (ablation) | EGNN(C) · Cora dense | accuracy: 85.1\pm{}0.5\% -> 88.8\pm{}0.3\% | Improved |
| Add edge adaptiveness across layers (ablation) | EGNN(A) · Cora dense | accuracy: 87.9\pm{}0.4\% -> 88.4\pm{}0.3\% | Improved |
| Add multidimensional directed-edge features and doubly stochastic normalization together (ablation) | EGNN(C) · Cora dense | accuracy: EGNN(C)-M 85.1\pm{}0.5\%; EGNN(C)-D 86.1\pm{}0.5\%; EGNN(C) 88.8\pm{}0.3\% | Improved |
| Use weighted loss (ablation) | EGNN models · Pubmed | accuracy | Mixed |
Closest papers by meaning
- EdgeNets: Edge Varying Graph Neural Networks (TPAMI 2021 · similarity 0.93)
- Co-Embedding of Nodes and Edges With Graph Neural Networks (TPAMI 2020 · similarity 0.90)
- Edge Prompt Tuning for Graph Neural Networks (ICLR 2025 · similarity 0.89)
- Improving Graph Neural Networks by Learning Continuous Edge Directions (ICLR 2025 · similarity 0.89)
- Towards characterizing the value of edge embeddings in Graph Neural Networks (ICML 2025 · similarity 0.88)
- Graph Attention is Not Always Beneficial: A Theoretical Analysis of Graph Attention Mechanisms via Contextual Stochastic Block Models (ICML 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.
