Research map / Graph learning / GNN architectures
Evaluating Post-hoc Explanations for Graph Neural Networks via Robustness Analysis
Junfeng Fang, Wei Liu, Yuan Gao, Zemin Liu, An Zhang, Xiang Wang, Xiangnan He
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 |
|---|---|---|---|
| use OAR instead of removal-based evaluation | explanation evaluation · BA3 | Kendall tau: 0.73 -> 1.00 | Improved |
| use OAR instead of DSE | explanation evaluation · MNIST-sp | Kendall tau: 0.20 -> 0.93 | Improved |
| use OAR instead of RM | explanation evaluation · MUTAG | Kendall tau: 0.73 -> 1.00 | Improved |
| use OAR instead of RM | explanation evaluation · TR3 | Kendall tau: 0.67 -> 0.93 | Improved |
| remove OOD reweighting block (OAR to SimOAR) (ablation) | OAR · BA3 | Kendall tau: 1.00 -> 0.93 | Hurt |
| remove OOD reweighting block (OAR to SimOAR) (ablation) | OAR · MUTAG | Kendall tau: 1.00 -> 0.80 | Hurt |
| remove OOD reweighting block (OAR to SimOAR) (ablation) | OAR · MNIST-sp | Kendall tau: 0.93 -> 0.93 | No clear effect |
| remove OOD reweighting block (ablation) | OAR · average over datasets | evaluation score: 7.83% less | Hurt |
| use SimOAR instead of RM | explanation evaluation · average over datasets | evaluation score: 42.62% higher | Improved |
| use SimOAR instead of DSE | explanation evaluation · average over datasets | evaluation score: 37.45% higher | Improved |
| use SimOAR instead of OAR for speed (ablation) | OAR · BA3 | time per graph (ms): 1.16 -> 0.84 | Improved |
| use SimOAR instead of OAR for speed (ablation) | OAR · MNIST-sp | time per graph (ms): 1.77 -> 1.03 | Improved |
Closest papers by meaning
- Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks (ICLR 2024 · similarity 0.92)
- Reconsidering Faithfulness in Regular, Self-Explainable and Domain Invariant GNNs (ICLR 2025 · similarity 0.91)
- Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 2024 · similarity 0.90)
- Graph Neural Network Explanations are Fragile (ICML 2024 · similarity 0.89)
- Redundancy Undermines the Trustworthiness of Self-Interpretable GNNs (ICML 2025 · similarity 0.89)
- GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking (ICLR 2024 · similarity 0.89)
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.
