Research map / Graph learning / GNN architectures
GNN explainability
Evaluate and generate explanations for GNN predictions.
64 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2019–2026) · an approach in GNN architectures, keyword cluster “gnn · explanations · trained”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
22 papers in 2023–24 and 28 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 4.6% to 5.5%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2019 · GNNExplainer: Generating Explanations for Graph Neural Networks (NeurIPS 2019 · 687 citations)
- 2020 · Parameterized Explainer for Graph Neural Network (NeurIPS 2020 · 210 citations)
- 2021 · On Explainability of Graph Neural Networks via Subgraph Explorations (ICML 2021 · 90 citations)
- 2022 · ProtGNN: Towards Self-Explaining Graph Neural Networks (AAAI 2022 · 116 citations)
- 2023 · Global Concept-Based Interpretability for Graph Neural Networks via Neuron Analysis (AAAI 2023 · 31 citations)
- 2024 · Factorized Explainer for Graph Neural Networks (AAAI 2024 · 9 citations)
- 2025 · On Logic-based Self-Explainable Graph Neural Networks (NeurIPS 2025 · 2 citations)
- 2026 · GNN Explanations that do not Explain and How to find Them (ICLR 2026 · 0 citations)
Earliest papers here
- GNNExplainer: Generating Explanations for Graph Neural Networks (NeurIPS 2019 · 687 citations)
- Parameterized Explainer for Graph Neural Network (NeurIPS 2020 · 210 citations)
- What Can Neural Networks Reason About? (ICLR 2020 · 97 citations)
What has been tried
46 papers here report 463 measured changes (a component added, removed or swapped, compared with a baseline): helped 66%, no clear effect 8%, hurt 20%, mixed 6%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Increase k from 1 to 60 | BA-Community, Tree-Circles, Tree-Grid, BA-2motifs, MUTAG · AUC-ROC | BA-Community: 0.850 -> 0.952; Tree-Circles: 0.758 -> 0.778; Tree-Grid: 0.711 -> 0.770; BA-2motifs: 0.580 -> 0.588; MUTAG | Helped | Factorized Explainer for Graph Neural Networks (AAAI 2024 · 9 citations) |
| reduce supervision to 5 labeled atoms per predicate | Hanoi · ROC-AUC | 99.99 -> 90.62 | Hurt | Relational Concept Bottleneck Models (NeurIPS 2024 · 5 citations) |
| use exact rather than sampling-based Shapley computation | BAMultiShapes · fidelity | 0.86 ± 0.01 -> 0.87 ± 0.02 | Helped | GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules (NeurIPS 2024 · 5 citations) |
| add contrastive loss Lcts | Citeseer · accuracy | 69.01±1.2 -> 71.26±0.7 | Helped | Self-Interpretable Graph Learning with Sufficient and Necessary Explanations (AAAI 2024 · 4 citations) |
| remove mix-up processing | all four datasets · AUC-ROC | Hurt | RegExplainer: Generating Explanations for Graph Neural Networks in Regression Tasks (NeurIPS 2024 · 4 citations) | |
| add attention-based motif identification instead of individual-motif scoring | Mutagenicity · Average Probability | 0.9892 -> 0.9959 | Helped | MAGE: Model-Level Graph Neural Networks Explanations via Motif-based Graph Generation (ICLR 2025 · 2 citations) |
| Add lifting of graphs to cell complexes | Synth-Wheel/Cube · GEA | Helped | Higher Order Structures for Graph Explanations (AAAI 2025 · 1 citation) | |
| Retrain the classifier with the frozen subgraph extractor | SPMotif 0.5 · accuracy | 46.85 -> 60.09 | Helped | How Interpretable Are Interpretable Graph Neural Networks? (ICML 2024 · 1 citation) |
| Add linear-complexity search to existing node- and edge-level explainers | MUTAG, BA-2Motifs, and Mutagenicity · overall fidelity | Helped | EiG-Search: Generating Edge-Induced Subgraphs for GNN Explanation in Linear Time (ICML 2024 · 1 citation) | |
| Use a larger perturbation budget | Attack performance across datasets · overlap ratio | Helped | Graph Neural Network Explanations are Fragile (ICML 2024 · 1 citation) | |
| remove the Rule channel | RedBlueNodes test · accuracy | \nentry{100}{00} -> \nentry{50}{02} | Hurt | Beyond Topological Self-Explainable GNNs: A Formal Explainability Perspective (ICML 2025 · 0 citations) |
| remove clustering | FCR / NCI1 · coverage | 42.9% -> 10.1% | Hurt | COMRECGC: Global Graph Counterfactual Explainer through Common Recourse (ICML 2025 · 0 citations) |
What reviewers will expect
Datasets and baselines named in the 20 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- Mutagenicity · 3 of 20 papers
- AIDS · 3 of 20 papers
- MUTAG · 2 of 20 papers
- BBBP · 2 of 20 papers
- Proteins · 2 of 20 papers
Baselines
- GNNExplainer · 5 of 20 papers
- SubgraphX · 3 of 20 papers
- PG-Explainer · 2 of 20 papers
- GSAT · 2 of 20 papers
- GNNInterpreter · 2 of 20 papers
- ProtGNN · 2 of 20 papers
- XGNN · 2 of 20 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 50 papers; from OpenAlex and the papers' affiliations).
- Florida International University · 5 papers
- Indian Institute of Technology · 3 papers
- Indian Institute of Technology Delhi · 3 papers
- University of Illinois at Chicago · 3 papers
- Arizona State University · 3 papers
- University of Cambridge · 2 papers
- University of Trento · 2 papers
- The Ohio State University · 2 papers
- University of Montreal · 2 papers
- University of Alberta · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- GNN generalization theory (187 papers)
- Simple and scalable GNNs (131 papers)
- Robustness, privacy, and fairness of GNNs (94 papers)
- Physics-inspired GNNs (81 papers)
- Graph structure learning (201 papers)
Elsewhere in the topic
- Expressive equivariant message passing (109 papers)
- Theory of graph convolutional networks (83 papers)
- Deep graph convolution and simplification (82 papers)
- Graph rewiring and scalable message passing (72 papers)
- Weisfeiler-Leman expressivity hierarchies (66 papers)
- Structural GNN variants (65 papers)
- Oversmoothing and heterophily (60 papers)
- Substructure-aware GNN expressivity (50 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).