令人惊叹的图可解释性论文

关于图神经网络可解释性的论文

综述类

1. [ACM计算综述25] 解释图神经网络中的解释器：一项比较研究 论文
2. [IEEE会议论文集24] 可信图神经网络：方面、方法与趋势 论文
3. [预印本24] 基于图的可解释AI：全面综述 论文
4. [Arxiv 23] 图神经网络可解释性综述 论文
5. [ACM计算综述] 图反事实解释综述：定义、方法、评估与研究挑战 论文
6. [TPAMI 22]图神经网络中的可解释性：分类学综述。袁浩、于海洋、桂树睿、季水旺。论文
7. [Arxiv 22]可解释图神经网络综述：分类与评估指标 论文
8. [Arxiv 22] 可信图学习综述：可靠性、可解释性与隐私保护 论文
9. [大数据2022]用于网络恶意软件分析的可解释图神经网络综述 论文
10. [机器智能研究24] 可信图神经网络综合综述：隐私、鲁棒性、公平性与可解释性论文
11. [书籍23] 图神经网络的生成式解释：方法与评估 论文

平台类

1. PyTorch Geometric [文档] [博客]
2. DIG：深入图深度学习研究的一站式库 论文 代码
3. GraphXAI：评估图神经网络的可解释性 论文 代码
4. GraphFramEx：迈向图神经网络可解释性方法的系统化评估 论文 代码
5. GNNExplainer和PGExplainer 论文 代码
6. BAGEL：评估图神经网络解释的基准测试 [论文]代码

由Cogdl选出的最具影响力论文

1. 图神经网络中的可解释性：分类学综述。袁浩、于海洋、桂树睿、季水旺。ARXIV 2020年。论文
2. Gnnexplainer：为图神经网络生成解释。Ying Rex、Bourgeois Dylan、You Jiaxuan、Zitnik Marinka、Leskovec Jure。NeurIPS 2019年。论文 代码
3. 图卷积神经网络的可解释性方法。Pope Phillip E、Kolouri Soheil、Rostami Mohammad、Martin Charles E、Hoffmann Heiko。CVPR 2019年。论文
4. 图神经网络的参数化解释器。Luo Dongsheng、Cheng Wei、Xu Dongkuan、Yu Wenchao、Zong Bo、Chen Haifeng、Zhang Xiang。NeurIPS 2020年。论文 代码
5. Xgnn：迈向图神经网络的模型级解释。袁浩、Tang Jiliang、Hu Xia、季水旺。KDD 2020年。论文。
6. 评估图神经网络的归因。Sanchez-Lengeling Benjamin、Wei Jennifer、Lee Brian、Reif Emily、Wang Peter、Qian Wesley、McCloskey Kevin、Colwell Lucy、Wiltschko Alexander。NeurIPS 2020年。论文
7. PGM-Explainer：图神经网络的概率图模型解释。Vu Minh、Thai My T.。NeurIPS 2020年。论文
8. 基于解释的弱监督视觉关系学习与图网络。Federico Baldassarre、Kevin Smith、Josephine Sullivan、Hossein Azizpour。ECCV 2020年。论文
9. GCAN：面向社交媒体上可解释假新闻检测的图感知协同注意力网络。Lu、Yi-Ju和Li、Cheng-Te。ACL 2020年。论文
10. 通过子图探索实现图神经网络可解释性。袁浩、于海洋、Wang Jie、Li Kang、季水旺。ICML 2021年。论文

2026年

• [ICLR 26] 无法真正解释的GNN解释及其发现方法[论文]

2025年

• [NIPS 25] 基于图曲率的图神经网络鲁棒解释[论文]
• [NIPS 25] GnnXemplar：示例即解释——用于全局图神经网络可解释性的自然语言规则[论文]
• [NIPS 25] 基于逻辑的自解释图神经网络研究[论文]

1. [ICLR 25] 图神经网络任意阶Shapley交互作用的精确计算[论文]
2. [ICLR 25] 从图神经网络到树结构：图神经网络的多粒度可解释性[论文]
3. [ICLR 25] 针对图扰动攻击的可证明鲁棒可解释图神经网络[论文]
4. [ICLR 25] 迈向解释常深度图神经网络在线性规划中的能力[论文]
5. [ICLR 25] 基于公理化层边的演化图上GNN解释[论文]
6. [ICLR 25] MAGE：基于基元图生成的模型级图神经网络解释[论文]
7. [AAAI 25] 用于图解释的高阶结构[论文]
8. [AAAI 25] 用于链接符号预测的自解释图Transformer[论文]
9. [AAAI 25] 基于计算树视角的忠实且准确的自注意力归因，应用于消息传递神经网络[论文]
10. [AAAI 25] 图分割与对比增强型图神经网络解释器[论文]
11. [TKDD 25] DyExplainer：可解释的动态图神经网络[论文]
12. [Arxiv 25.05] 基于子图匹配的双重解释方法用于恶意软件检测[论文]
13. [Arxiv 25.04] 关于图神经网络在恶意软件检测中解释的一致性问题[论文]
14. [Arxiv 25.01] 通过解释技术对图神经网络进行水印标记以保护所有权[论文]
15. [Arxiv 25.01] 混合专家图Transformer用于可解释的粒子碰撞检测[论文]
16. [ACM Computing Surveys] 图神经网络能否被充分解释？一项综述[论文]
17. [IEEE TNSRE] 利用可解释图神经网络寻找运动学习与康复的神经生物标志物[论文]
18. [Springer FCS] 借鉴捷径：一种基于捷径引导的可解释图学习方法[论文]
19. [NN] 具有多头图通道注意力网络的局部可解释垃圾信息检测模型[论文]
20. [Applied Intelligence] KnowGNN：一种知识感知、结构敏感的模型级图神经网络解释器[论文]
21. [ICML 25] TopInG：基于持久性理由过滤的拓扑可解释图学习[论文] [项目]

Year 2024

1. [NeurIPS 24] RegExplainer: Generating Explanations for Graph Neural Networks in Regression Task [paper]
2. [NeurIPS 24] GraphTrail: Translating GNN Predictions into Human-Interpretable Logical Rules[paper]
3. [ICML 24] Generating In-Distribution Proxy Graphs for Explaining Graph Neural Networks[paper]
4. [ICML 24] Predicting and Interpreting Energy Barriers of Metallic Glasses with Graph Neural Networks [paper]
5. [ICML 24] Graph Neural Network Explanations are Fragile [paper]
6. [ICML 24] How Interpretable Are Interpretable Graph Neural Networks? [paper]
7. [ICML 24] Feature Attribution with Necessity and Sufficiency via Dual-stage Perturbation Test for Causal Explanation[paper]
8. [ICML 24] Explaining Graph Neural Networks via Structure-aware Interaction Index [paper]
9. [ICML 24] EiG-Search: Generating Edge-Induced Subgraphs for GNN Explanation in Linear Time [paper]
10. [ICLR 24] GraphChef: Decision-Tree Recipes to Explain Graph Neural Networks [paper]
11. [ICLR 24] GOAt: Explaining Graph Neural Networks via Graph Output Attribution [paper]
12. [ICLR 24] Towards Robust Fidelity for Evaluating Explainability of Graph Neural Networks [paper]
13. [ICLR 24] GNNX-BENCH: Unravelling the Utility of Perturbation-based GNN Explainers through In-depth Benchmarking [paper]
14. [ICLR 24] UNR-Explainer: Counterfactual Explanations for Unsupervised Node Representation Learning Models [paper]
15. [TPAMI 24] Towards Inductive and Efficient Explanations for Graph Neural Networks[paper]
16. [Openreview 24] Robust Graph Attention for Graph Adversarial Attacks: An Information Bottleneck Inspired Approach[paper]
17. [Openreview 24] AIMing for Explainability in GNNs[paper]
18. [Openreview 24] Graph Distributional Analytics: Enhancing GNN Explainability through Scalable Embedding and Distribution Analysis[paper]
19. [Openreview 24] Watermarking Graph Neural Networks Via Explanations For Ownership Protection[paper]
20. [Openreview 24] Explainable Graph Representation Learning via Graph Pattern Analysis [paper]
21. [Openreview 24] Robust Heterogeneous Graph Neural Network Explainer with Graph Information Bottleneck [paper]
22. [Openreview 24] A Hierarchical Language Model Design For Interpretable Graph Reasoning [paper]
23. [Openreview 24] The GECo algorithm for Graph Neural Networks Explanation [paper]
24. [Openreview 24] On Explaining Equivariant Graph Networks via Improved Relevance Propagation [paper]
25. [Openreview 24] SIG: Self-Interpretable Graph Neural Network for Continuous-time Dynamic Graphs [paper]
26. [Openreview 24] Interpretable and Adaptive Graph Contrastive Learning with Information Sharing for Biomedical Link Prediction [paper]
27. [Openreview 24] TAGExplainer: Narrating Graph Explanations for Text-Attributed Graph Learning Models [paper]
28. [Openreview 24] TreeX: Generating Global Graphical GNN Explanations via Critical Subtree Extraction [paper]
29. [TMLR 24] InduCE: Inductive Counterfactual Explanations for Graph Neural Networks [paper]
30. [PLDI 24] PL4XGL: A Programming Language Approach to Explainable Graph Learning[paper]
31. [Usenix Security 24] INSIGHT: Attacking Industry-Adopted Learning Resilient Logic Locking Techniques Using Explainable Graph Neural Network[paper]
32. [SIGMOD 24]View-based Explanations for Graph Neural Networks [paper]
33. [ACM SIGMOD Record] The Road to Explainable Graph Neural Networks [paper]
34. [Thesis UCLA] Explainable Artificial Intelligence for Graph Data[paper]
35. [Thesis UVA] Algorithmic Fairness in Graph Machine Learning: Explanation, Optimization, and Certification[paper]
36. [KDD 24] SEFraud: Graph-based Self-Explainable Fraud Detection via Interpretative Mask Learning[paper]
37. [KDD 24] Self-Explainable Temporal Graph Networks based on Graph Information Bottleneck[paper]
38. [KDD 24] Unveiling Global Interactive Patterns across Graphs: Towards Interpretable Graph Neural Networks[paper]
39. [ICDE 24] Generating Robust Counterfactual Witnesses for Graph Neural Networks [paper]
40. [ICDE 24] SES: Bridging the Gap Between Explainability and Prediction of Graph Neural Networks[paper]
41. [ICSE 24] Coca: Improving and Explaining Graph Neural Network-Based Vulnerability Detection Systems[paper]
42. [AAAI 24] Generating Diagnostic and Actionable Explanations for Fair Graph Neural Networks [paper]
43. [AAAI 24] Stratifed GNN Explanations through Sufficient Expansion[paper]
44. [AAAI 24] Factorized Explainer for Graph Neural Networks[paper]
45. [AAAI 24] Self-Interpretable Graph Learning with Sufficient and Necessary Explanations
46. [AAAI 24] Explainable Origin-Destination Crowd Flow Interpolation via Variational Multi-Modal Recurrent Graph Auto-Encoder [paper]
47. [AISTATS 24] Two Birds with One Stone: Enhancing Uncertainty Quantification and Interpretability with Graph Functional Neural Process [paper]
48. [WWW 24] Game-theoretic Counterfactual Explanation for Graph Neural Networks [paper]
49. [WWW 24] EXGC: Bridging Efficiency and Explainability in Graph Condensation[paper]
50. [WWW 24] Adversarial Mask Explainer for Graph Neural Networks [paper]
51. [WWW 24] Globally Interpretable Graph Learning via Distribution Matching[paper]
52. [WWW 24] GNNShap: Scalable and Accurate GNN Explanation using Shapley Values [paper]
53. [LOG 24] xAI-Drop: Don't Use What You Cannot Explain[paper]
54. [LOG 24] MOSE-GNN: A Motif-Based Self-Explaining Graph Neural Network for Molecular Property Prediction [paper]
55. [TNNLS 24] BrainIB: Interpretable Brain Network-based Psychiatric Diagnosis with Graph Information Bottleneck [paper]
56. [TKDE 24] On Regularization for Explaining Graph Neural Networks: An Information Theory Perspective [paper]
57. [TKDD 24] Towards Prototype-Based Self-Explainable Graph Neural Network [paper]
58. [TKDD 24] Efficient GNN Explanation via Learning Removal-based Attribution [paper]
59. [TAI 24] Learning Counterfactual Explanation of Graph Neural Networks via Generative Flow Network[paper]
60. [TAI 24] Traffexplainer: A Framework towards GNN-based Interpretable Traffic Prediction [paper]
61. [TMC 24] HGExplainer: Heterogeneous Graph Explainer for IoT Device Identification[paper]
62. [IEEE TMI 24] Multi-Modal Diagnosis of Alzheimer’s Disease using Interpretable Graph Convolutional Networks[paper]
63. [IEEE IoT 24] EXVul: Toward Effective and Explainable Vulnerability Detection for IoT Devices[paper]
64. [IEEE Transactions on Fuzzy Systems] Towards Embedding Ambiguity-Sensitive Graph Neural Network Explainability [paper]
65. [IEEE JBHI] Interpretable Dynamic Directed Graph Convolutional Network for Multi-Relational Prediction of Missense Mutation and Drug Response[paper]
66. [IDEAL 2024] Causal Explanation of Graph Neural Networks[paper]
67. [BIBM 24] Seizure Onset Zone Localization Method based on GNN Explanation [paper]
68. [BIBM 24] DDTExplainer: Mining Drug-Disease Therapeutic Mechanisms based on GNN Explainability [paper]
69. [CIKM 24] EDGE: Evaluation Framework for Logical vs. Subgraph Explanations for Node Classifiers on Knowledge Graphs[paper]
70. [ECML/PKDD 24] Towards Few-shot Self-explaining Graph Neural Networks[paper]
71. [SDM 24] XGExplainer: Robust Evaluation-based Explanation for Graph Neural Networks[paper]
72. [DASFAA 24] Multi-objective Graph Neural Network Explanatory Model with Local and Global Information Preservation[paper]
73. [ISSTA 2024] Graph Neural Networks for Vulnerability Detection: A Counterfactual Explanation [paper]
74. [KBS 24] Shapley-based graph explanation in embedding space[paper]
75. [KBS 24] GEAR: Learning graph neural network explainer via adjusting gradients[paper]
76. [IEEE TNSM 24] Ensemble Graph Attention Networks for Cellular Network Analytics: From Model Creation to Explainability[paper]
77. [IEEE TNSE 24] GAXG: A Global and Self-adaptive Optimal Graph Topology Generation Framework for Explaining Graph Neural Networks[paper]
78. [IEEE TETCI 24] GF-LRP: A Method for Explaining Predictions Made by Variational Graph Auto-Encoders[paper]
79. [AAAI workshop] Semi-Supervised Graph Representation Learning with Human-centric Explanation for Predicting Fatty Liver Disease[paper]
80. [xAI 24] Global Concept Explanations for Graphs by Contrastive Learning [paper]
81. [Arxiv 24.12] BetaExplainer: A Probabilistic Method to Explain Graph Neural Networks[paper]
82. [Arxiv 24.12] GISExplainer: On Explainability of Graph Neural Networks via Game-theoretic Interaction Subgraphs [paper]
83. [Arxiv 24.12] Interpreting GNN-based IDS Detections Using Provenance Graph Structural Features [paper]
84. [Arxiv 24.12] eXpath: Explaining Knowledge Graph Link Prediction with Ontological Closed Path Rules[paper]
85. [Arxiv 24.12] On the Probability of Necessity and Sufficiency of Explaining Graph Neural Networks: A Lower Bound Optimization Approach [paper]
86. [Arxiv 24.11] Rethinking Node Representation Interpretation through Relation Coherence[paper]
87. [Arxiv 24.11] MBExplainer: Multilevel bandit-based explanations for downstream models with augmented graph embeddings [paper]
88. [Arxiv 24.11] Securing GNNs: Explanation-Based Identification of Backdoored Training Graphs[paper]
89. [Preprint 24.11] Chiseling the Graph: An Edge-Sculpting Method for Explaining Graph Neural Networks [paper]
90. [Preprint 24.10] Reliable and Faithful Generative Explainers for Graph Neural Networks[paper]
91. [Arxiv 24.10] Explaining Hypergraph Neural Networks: From Local Explanations to Global Concepts[paper]
92. [Arxiv 24.10] Explainable Graph Neural Networks Under Fire [paper]
93. [Arxiv 24.09] GINTRIP: Interpretable Temporal Graph Regression using Information bottleneck and Prototype-based method [paper]
94. [Arxiv 24.09] PAGE: Parametric Generative Explainer for Graph Neural Network [paper]
95. [Preprint 24.08] CIDER: Counterfactual-Invariant Diffusion-based GNN Explainer for Causal Subgraph Inference[paper]
96. [Arxiv 24.07] LLMExplainer: Large Language Model based Bayesian Inference for Graph Explanation Generation[paper]
97. [Arxiv 24.07] Explaining Graph Neural Networks for Node Similarity on Graphs[paper]
98. [Arxiv 24.07] SLInterpreter: An Exploratory and Iterative Human-AI Collaborative System for GNN-based Synthetic Lethal Prediction[paper]
99. [Arxiv 24.07] Graph Neural Network Causal Explanation via Neural Causal Models[paper]
100. [Arxiv 24.06] GNNAnatomy: Systematic Generation and Evaluation of Multi-Level Explanations for Graph Neural Networks[paper]
101. [Arxiv 24.06] On GNN explanability with activation rules[paper]
102. [Arxiv 24.05] SIG: Efficient Self-Interpretable Graph Neural Network for Continuous-time Dynamic Graphs[paper]
103. [Arxiv 24.06] L2XGNN: Learning to Explain Graph Neural Networks [paper]
104. [Arxiv 24.06] Towards Understanding Sensitive and Decisive Patterns in Explainable AI: A Case Study of Model Interpretation in Geometric Deep Learning[paper]
105. [Arxiv 24.06] Explainable AI Security: Exploring Robustness of Graph Neural Networks to Adversarial Attacks [paper]
106. [Arxiv 24.06] Robust Ante-hoc Graph Explainer using Bilevel Optimization [paper]
107. [Arxiv 24.06] Perks and Pitfalls of Faithfulness in Regular, Self-Explainable and Domain Invariant GNNs [paper]
108. [Arxiv 24.05] Utilizing Description Logics for Global Explanations of Heterogeneous Graph Neural Networks [paper]
109. [Arxiv 24.05] Detecting Complex Multi-step Attacks with Explainable Graph Neural Network [paper]
110. [Arxiv 24.05] SynHING: Synthetic Heterogeneous Information Network Generation for Graph Learning and Explanation[paper]
111. [Arxiv 24.05] PAGE: Prototype-Based Model-Level Explanations for Graph Neural Networks [paper]
112. [Arxiv 24.05] Evaluating Neighbor Explainability for Graph Neural Networks [paper]
113. [Preprint 24.05] Explainable Graph Neural Networks: An Application to Open Statistics Knowledge Graphs for Estimating House Prices [paper]
114. [Arxiv 24.04] Superior Polymeric Gas Separation Membrane Designed by Explainable Graph Machine Learning [paper]
115. [Arxiv 24.04] Improving the interpretability of GNN predictions through conformal-based graph sparsification [paper]
116. [Arxiv 24.03] GreeDy and CoDy: Counterfactual Explainers for Dynamic Graph[paper]
117. [Arxiv 24.03] Explainable Graph Neural Networks for Observation Impact Analysis in Atmospheric State Estimation[paper]
118. [Arixv 24.03] Iterative Graph Neural Network Enhancement via Frequent Subgraph Mining of Explanations[paper]
119. [Arxiv 24.02] PAC Learnability under Explanation-Preserving Graph Perturbations[paper]
120. [Arxiv 24.02] Explainable Global Wildfire Prediction Models using Graph Neural Networks[paper]
121. [Arxiv 24.02] Incorporating Retrieval-based Causal Learning with Information Bottlenecks for Interpretable Graph Neural Networks[paper]
122. [Arxiv 24.01] On Discprecncies between Perturbation Evaluations of Graph Neural Network Attributions[paper]
123. [ASP=DAC 24] LIPSTICK: Corruptibility-Aware and Explainable Graph Neural Network-based Oracle-Less Attack on Logic Locking[paper]
124. [Biorxiv 24] Community-aware explanations in knowledge graphs with XP-GNN[paper]
125. [ISCV 24] Adaptive Subgraph Feature Extraction for Explainable Multi-Modal Learning[paper]
126. [IJCNN 24] Explanations for Graph Neural Networks using A Game-theoretic Value[paper]
127. [AIxIA 2024] Relating Explanations with the Inductive Biases of Deep Graph Networks [paper]
128. [Neurocomputing] GeoExplainer: Interpreting Graph Convolutional Networks with geometric masking[paper]
129. [Technologies] Explainable Graph Neural Networks: An Application to Open Statistics Knowledge Graphs for Estimating House Prices[paper]
130. [Reliab. Eng. Syst. Saf.] Causal intervention graph neural network for fault diagnosis of complex industrial processes[paper]
131. [Frontiers in big data] Global explanation supervision for Graph Neural Networks[paper]
132. [Information and Software Technology] Graph-based explainable vulnerability prediction[paper]
133. [Information Systems] Heterogeneous graph neural networks for fraud detection and explanation in supply chain finance[paper]
134. [Information Procs. & Mana.] Towards explaining graph neural networks via preserving prediction ranking and structural dependency[paper]
135. [Applied Energy] Explainable Spatio-Temporal Graph Neural Networks for multi-site photovoltaic energy production [paper]
136. [PAKDD 24] Random Mask Perturbation Based Explainable Method of Graph Neural Networks [paper]
137. [Computational Materials Science] Graph isomorphism network for materials property prediction along with explainability analysis[paper]
138. [NN 24] Explanatory subgraph attacks against Graph Neural Networks[paper]
139. [NN 24] GRAM: An interpretable approach for graph anomaly detection using gradient attention maps[paper]
140. [Neural Networks 24] CI-GNN: A Granger Causality-Inspired Graph Neural Network for Interpretable Brain Network-Based Psychiatric Diagnosis [paper]
141. [NeuroImage 24] BPI-GNN: Interpretable brain network-based psychiatric diagnosis and subtyping[paper]
142. [PAKDD 24] Toward Interpretable Graph Classification via Concept-Focused Structural Correspondence [paper]
143. [ICPR 24] Interpretable Deep Graph-Level Clustering: A Prototype-Based Approach [paper]
144. [MedRxiv 24] An Interpretable Population Graph Network to Identify Rapid Progression of Alzheimer’s Disease Using UK Biobank[paper]
145. [IEEE TDSC 24] TrustGuard: GNN-based Robust and Explainable Trust Evaluation with Dynamicity Support [paper]
146. [IEEE Transactions] IEEE Transactions on Computational Social Systems[paper]
147. [Journal of Physics] Explainer on GNN-based segmentation networks[paper]
148. [Energy and AI] Electricity demand forecasting at distribution and household levels using explainable causal graph neural network [paper]
149. [HI-AI@KDD 24] Interpretable Graph Model with Prototype-Based Graph Information Bottleneck [paper]
150. [Neurosymbolic Artificial Intelligence] Towards Semantic Understanding of GNN Layers embedding with Functional-Semantic Activation Mapping [paper]
151. [NeSy 2024] Towards Understanding Graph Neural Networks: Functional-Semantic Activation Mapping[paper]
152. [Thesis 24] Explainable and physics-guided graph deep learning for air pollution modelling [paper]
153. [Thesis 24] Influence of molecular structures on graph neural network explainers’ performance[paper]

2023年

1. [NeurIPS 23] 可解释图网络提出通用代数猜想[论文]
2. [NeurIPS 23] SAME：基于结构感知Shapley值的分段式解释方法揭示GNN黑箱[论文]
3. [NeurIPS 23] 一次训练，处处解释：预训练可解释图神经网络[论文]
4. [NeurIPS 23] D4Explainer：通过离散去噪扩散模型实现图神经网络的分布内解释[论文]
5. [NeurIPS 23] TempME：基于模体发现提升时序图神经网络的可解释性[论文]
6. [NeurIPS 23] V-InFoR：针对结构受损图的鲁棒图神经网络解释器[论文]
7. [NeurIPS 23] 迈向自解释的图级别异常检测[论文]
8. [NeurIPS 23] 基于鲁棒性分析评估图神经网络的事后解释方法[论文]
9. [NeurIPS 23] 基于原型的可解释图信息瓶颈[论文]
10. [ICML 23] 重新思考：通过非参数子图匹配解释图神经网络[论文]
11. [ICML 23] 相关游走搜索：用于解释图神经网络的方法[论文]
12. [ICML 23] 迈向理解图神经网络的泛化能力[论文]
13. [ICLR 23] GNNInterpreter：面向图神经网络的概率生成模型级解释[论文]
14. [ICLR 23] 通过学习概念的逻辑组合实现GNN的全局可解释性[论文]
15. [ICLR 23] 基于探索-导航框架解释时序图模型[论文]
16. [ICLR 23] DAG很重要！增强版GFlowNets用于图神经网络的解释[论文]
17. [ICLR 23] 通过可学习的随机性注入实现可解释的几何深度学习[论文]
18. [ICLR 23] 从微分几何视角看演化图上的GNN及其可解释性[论文]
19. [KDD 23] MixupExplainer：利用数据增强推广图神经网络的解释方法[论文]
20. [KDD 23] 基于贪婪扰动的异质图反事实学习[论文]
21. [KDD 23] 借助信息瓶颈赋能事后图解释：预训练与微调视角[论文]
22. [KDD 23] 少即是多：SlimG用于精准、鲁棒且可解释的图挖掘[论文]
23. [KDD 23] 基于因果子结构的分子关系学习具有迁移鲁棒性[论文]
24. [AAAI 23] 基于神经元分析实现图神经网络的全局概念级可解释性[论文]
25. [AAAI 23] 关于解释黑盒时序图神经网络的极限[论文]
26. [AAAI 23] 迈向异质图神经网络的细粒度可解释性[论文]
27. [AAAI 23] 可解释的手性感知图神经网络用于药物发现中的定量构效关系建模[论文]
28. [VLDB 23] HENCE-X：面向异质性的多层级深度图网络解释框架[论文]
29. [VLDB 23] 关于图神经网络的数据感知型全局可解释性[论文]
30. [AISTATS 23] 蒸馏并解释：使用简单代理模型解释图神经网络[论文]
31. [AISTATS 23] 探究图表示[论文]
32. [ICDE 23] INGREX：图神经网络的交互式解释框架[论文]
33. [ICDE 23] 联合攻击图神经网络及其解释方法[论文]
34. [WWW 23] PaGE-Link：基于路径的异质链接预测图神经网络解释[论文]
35. [ICDM 23] 基于扰动的解释方法在时序图神经网络中的局限性
36. [ICDM 23] 面向超链接预测的可解释子图特征提取[论文]
37. [WSDM 23] 利用时序异质图检测研究兴趣变化的可解释方法[论文]
38. [WSDM 23] 图神经网络的合作式解释[论文]
39. [WSDM 23] 迈向忠实且一致的图神经网络解释[论文]
40. [WSDM 23] 图神经网络的全局反事实解释器[论文]
41. [CIKM 23] 可解释的时空图神经网络[论文]
42. [CIKM 23] DuoGAT：双时间导向的图注意力网络，用于准确、高效且可解释的时序异常检测[论文]
43. [CIKM 23] 异质时序图神经网络解释器[论文]
44. [CIKM 23] ACGAN-GNNExplainer：辅助条件生成式图神经网络解释器[论文]
45. [CIKM 23] KG4Ex：基于可解释知识图谱的运动推荐方法[论文]
46. [ECML-PKDD 23] ENGAGE：基于解释引导的数据增强用于图表示学习[论文]
47. [TPAMI 23] FlowX：通过消息流实现可解释的图神经网络[论文]
48. [TAI] 基于原型的可解释图神经网络[论文]
49. [TKDE 23] 基于反事实图学习的属性图异常检测[论文]
50. [Scientific Data 23] 评估图神经网络的可解释性[论文]
51. [Nature Communications 23] 结合子结构掩码的化学直观解释用于分子性质预测的图神经网络[论文]
52. [ACM Computing Surveys 23] 图反事实解释综述：定义、方法与评估[论文]
53. [TIST 23] 基于理由对齐的忠实且一致的图神经网络解释[论文]
54. [Openreview 23] STExplainer：通过频繁子树挖掘实现GNN的全局可解释性[论文]
55. [GLFrontiers 23] 人人都需要一点帮助：基于层次化概念解释图[论文]
56. [Openreview 23] 利用解释迭代增强图神经网络[论文]
57. [Openreview 23] 大规模可解释且收敛的图神经网络层[论文]
58. [NeurIPS 2023 Workshop XAIA] GInX-Eval：迈向图神经网络解释的分布内评估[论文]
59. [NeurIPS 2023 Workshop XAIA] 关于GNN解释方法的一致性[论文]
60. [Arxiv 23] 基于可解释性的图对抗攻击：通过边扰动生成对抗样本[论文]
61. [AICS 23] 基于单关系转换的知识图链接预测的子图解释生成模型[论文]
62. [GUT 23] 利用可解释图学习筛查正常内镜下结直肠活检：一项回顾性研究[论文]
63. [PR 2023] 迈向频率自适应Inception模块的自解释卷积图神经网络[论文]
64. [MLG 2023] 理解图神经网络中解释器的工作机制[论文]
65. [MLG 2023] 图模型解释工具[论文]
66. [Information Science 23] 带有神经元解释组件的图神经网络鲁棒解释[论文]
67. [Recsys 23] 可解释的图神经网络推荐系统：挑战与机遇[论文]
68. [xAI 23] 通过密度视角解释图分类的反事实解释[论文]
69. [XAI 23] 评估图神经网络的链接预测解释[论文]
70. [XAI 23] XInsight：基于流式解释揭示GNN的模型洞察[论文]
71. [XAI 23] 通过人工可模拟性研究量化图神经网络归因式解释的内在效用[论文]
72. [XAI 23] MEGAN：多解释图注意力网络[论文]
73. [XKDD 23] 图神经网络的博弈论解释[论文]
74. [XKDD 23] 从黑箱到玻璃箱：评估GCNN过程预测的忠实性[论文]
75. [IJCNN 23] MEGA：利用网络模体解释图神经网络[论文]
76. [LOG Poster 23] 事后GNN解释器对标签噪声的鲁棒性[论文]
77. [LOG Poster 23] 自解释GNN有多忠实？[论文]
78. [LOG Poster 23] 用有影响力示例解释知识图嵌入模型中的链接预测[论文]
79. [Bioriv 23] 通过稀疏学习构建可解释的图神经网络用于药物-蛋白质结合预测[论文]
80. [ICAID 2023] 通过层分析解释图神经网络[论文]
81. [ECAI 23] XGBD：基于解释引导的图后门检测[论文]
82. [IEEE Transactions on Consumer Electronics 23] 利用可解释卷积图神经网络进行智能家居人体姿态预测[论文]
83. [KBS 23] KE-X：基于知识信息增益的知识图嵌入子图解释[论文]
84. [ICML workshop 23] 在领域约束下为分子生成全局事实与反事实解释器[论文]
85. [毕业论文 23] 开发高维特征空间下的可解释图神经网络[论文]
86. [毕业论文 23] 评估图神经网络在单细胞分类任务中的可解释性方法[论文]
87. [Arxiv 23] 关于子集选择与知情图神经网络的相互作用[论文]
88. [ISSTA23] 基于GNN的漏洞检测解释器：我们到了吗？[论文]
89. [ICECAI23] 基于交叉熵改进GNN解释的GraphSVX[论文]
90. [ICRA Workshop 23] 迈向语义解释与验证基于图注意力的解释[论文]
91. [Arxiv 23] 基于图神经网络的日志异常检测与解释[论文]
92. [毕业论文 23] 图形模型的可解释性[论文]
93. [生物工程 2023] 利用可解释图神经网络和人群图进行阿尔茨海默病早期诊断的个性化解释[论文]
94. [BDSC 2023] MDC：一种基于节点模体度和图扩散卷积的可解释GNN方法[论文]
95. [信息科学 2023] 将可解释技术应用于基于图神经网络模型的道路交通预测[论文]
96. [Arxiv 23.05] EiX-GNN：图神经网络的概念级特征中心性解释器[论文]
97. [Arxiv 23.04] 基于医学概念的图神经网络认知解释器[论文]
98. [ICLR Tiny 23] 消息传递选择：迈向可解释的图分类GNN[论文]
99. [ICLR Tiny 23] 重温CounteRGAN用于图的反事实可解释性[论文]
100. [MICCAI Workshop 23] IA-GCN：用于疾病预测的可解释注意力导向图卷积网络[论文]
101. [GRADES & NDA'23] 图神经网络可解释性方法演示[论文]
102. [Arxiv 23] 用于链接预测的自解释图神经网络[论文]
103. [Arxiv 23.02] MotifExplainer：基于模体的图神经网络解释器[论文]
104. [ChemRxiv 23] 利用梅尔森值解释化学信息学领域的图神经网络[论文]
105. [Neural Networks 23] 通过识别桥接节点的重要节点为Skip-gram基点嵌入生成事后解释[论文]
106. [ICASSP 23] 迈向更稳定和通用的子图信息瓶颈[论文]
107. [ESANN 23] 将随机解释器与子图神经网络结合可以提高表达能力和可解释性[论文]
108. [IEEE Access] 通过多专业学习者和在线知识蒸馏实时生成GNN解释[论文]
109. [IEEE Access] 通过归纳一致性预测为节点表示学习模型提供事后解释[论文]
110. [Journal of Software 23] 基于图神经网络的切片级漏洞检测与解释方法[论文]
111. [Automation in Construction 23] 从可解释的数据驱动隧道图中学习：用于堵塞检测的时空图卷积网络[论文]
112. [Briefings in Bioinformatics] 基于可解释图神经网络和多步聚焦机制预测分子性质[论文]
113. [Briefings in Bioinformatics] 通过图神经网络可视化识别关键化学信息[论文]
114. [Bioinformatics 23] 可解释的多层图神经网络用于癌症基因预测[论文]
115. [ICLR Workshop 23] GCI：图概念解释框架[论文]
116. [Arxiv 23] 利用HSIC为图神经网络提供结构化解释[论文]
117. [Internet of Things 23] XG-BoT：用于僵尸网络检测与取证的可解释深度图神经网络[论文]
118. [JOS23] 基于图神经网络的恶意软件检测通用解释与定位方法[论文]
119. [IJCNN 23] GRAPHSHAP：通过模体语言解释身份感知图分类器[论文]
120. [Arxiv 23.01] 子图增强型图神经网络中的可解释性[论文]

2022年

1. [NeurIPS 22] GStarX：基于通信结构感知的合作博弈的图级预测解释 [论文]
2. [NeurIPS 22] 通过学习解耦的因果子结构来去偏置图神经网络 [论文]
3. [NeurIPS 22] 任务无关的图神经网络解释 [论文]
4. [NeurIPS 22] CLEAR：图上的生成式反事实解释[论文]
5. [ICML 22] 基于随机注意力机制的可解释且可泛化的图学习 [论文]
6. [ICLR 22] DEGREE：面向图神经网络的分解式解释 [论文]
7. [ICLR 22] 知识图谱上可解释的图神经网络模型 [论文]
8. [ICLR 22] 发现图神经网络的不变性推理依据 [论文]
9. [KDD 22] 关于图神经网络中偏差的结构性解释 [论文]
10. [KDD 22] 用于可解释和可泛化图分类的因果注意力机制 [论文]
11. [CVPR 22] OrphicX：一种受因果启发的潜在变量模型，用于解释图神经网络 [论文]
12. [CVPR 22] 利用变分图信息瓶颈改进子图识别 [论文]
13. [AISTATS 22] 探查GNN解释器：对GNN解释方法的严格理论与实验分析 [论文]
14. [AISTATS 22] CF-GNNExplainer：图神经网络的反事实解释 [论文]
15. [TPAMI 22] 面向全图分类的差分隐私图神经网络 [论文]
16. [TPAMI 22] 面向图神经网络的强化因果解释器 [论文]
17. [VLDB 22] xFraud：异构图上的可解释欺诈交易检测 [论文]
18. [LOG 22]GraphFramEx：迈向图神经网络解释方法的系统性评估 [论文]
19. [LOG 22] 基于解释引导的消息传递训练图神经网络 [论文]
20. [The Webconf 22] 基于反事实与事实推理学习和评估图神经网络解释 [论文]
21. [AAAI 22] 基于原型的图神经网络解释 [论文]
22. [AAAI 22] KerGNNs：基于图核的可解释图神经网络[论文]
23. [AAAI 22] ProtGNN：迈向自解释的图神经网络 [论文]
24. [IEEE Big Data 22] 以公平性和权衡解释为代价降低图神经网络的准确性 [论文]
25. [CIKM 22] GRETEL：统一的图反事实解释评估框架 [论文]
26. [CIKM 22] GRETEL：图反事实解释评估框架[论文]
27. [CIKM 22] 面向图神经网络节点分类的模型中心型解释器 [论文]
28. [IJCAI 22] 我的GNN到底捕捉了什么？关于探索GNN内部表示的研究 [论文]
29. [ECML PKDD 22] 提升基于规则的GNN解释质量 [论文]
30. [MICCAI 22] 用于基于连接组学的大脑疾病分析的可解释图神经网络 [论文]
31. [MICCAI 22] 面向阿尔茨海默病多模态诊断的图卷积网络稀疏解释 [论文]
32. [EuroS&P 22] Illuminati：迈向用于网络安全分析的图神经网络解释 [论文]
33. [INFOCOM 22] 基于图神经网络的僵尸网络检测模型的可解释性评估 [论文]
34. [GLOBECOM 22] Shapley解释器——一种用于SDN中的GNN解释方法 [论文]
35. [GLOBECOM 22] 面向时序图神经网络的解释器 [论文]
36. [TKDE 22] Zorro：图神经网络中有效、稀疏且稳定的解释 [论文]
37. [TNNLS 22] 带有时间模式注意力的可解释图储备计算 [论文]
38. [TNNLS22] 用于训练可解释图神经网络的元学习方法 [论文]
39. [TNNLS 22] 通过输入扰动解释深度图网络 [论文]
40. [TNNLS 22] 用于训练可解释图神经网络的元学习方法 [论文]
41. [DMKD 22] 关于基于激活模式的GNN可解释性 [论文]
42. [KBS 22] EGNN：通过知识蒸馏构建可解释的图神经网络 [论文]
43. [XKDD 22] GREASE：为基于GNN的推荐系统生成事实与反事实解释 [论文]
44. [AI 22] 图神经网络解释器对图噪声是否鲁棒？ [论文]
45. [BRACIS 22] 面向图神经网络的ConveXplainer [论文]
46. [GLB 22] 用于基准测试图解释器的可解释AI库 [论文]
47. [DASFAA 22] 关于图神经网络的全局可解释性 [论文]
48. [ISBI 22] 用于阿尔茨海默病诊断的多模态脑成像可解释图卷积网络 [论文]
49. [生物信息学] GNN-SubNet：利用可解释图神经网络检测疾病亚网络 [论文]
50. [医学影像2022] 表型指导的可解释图卷积网络分析fMRI数据揭示青少年时期大脑连接的变化 [论文]
51. [NeuroComputing 22] 多扰动、多捕获：理解图神经网络的行为 [论文]
52. [DSN 22] CFGExplainer：从控制流图解释基于图神经网络的恶意软件分类 [论文]
53. [IEEE Access 22] 通过强化学习为node2vec提供节点级局部解释 [论文]
54. [Patterns 22] 面向分子性质预测的可解释图神经网络的定量评估 [论文]
55. [IEEE Access 22] 通过归纳一致性预测为节点表示学习模型提供事后解释 [论文]
56. [IEEE 22] 利用拓扑感知的节点选择解释图神经网络：应用于空气质量推断 [论文]
57. [BioRxiv 22] GNN-SubNet：利用可解释图神经网络检测疾病亚网络 [论文]
58. [IEEE Robotics and Automation Letters 22] 高效且可解释的机器人操作：基于图神经网络 [论文]
59. [Arxiv 22] 图神经网络中去混杂到解释评估 [论文]
60. [ICCPR 22] GANExplainer：基于GAN的图神经网络解释器 [论文]
61. [Arxiv 22] 探索图神经网络的解释方法 [论文]
62. [Arxiv 22] 通过在线知识蒸馏培养多领域学习者以解释GNN [论文]
63. [Openreview 23] TGP：面向个性化推荐的可解释时序图神经网络 [论文]
64. [Arxiv 22] PGX：基于独立知识蒸馏过程的多层级GNN解释框架 [论文]
65. [Arxiv 22] 通过可解释性防御图神经网络后门攻击 [论文]
66. [Arxiv 22] 通过相关性反向传播解释动态图神经网络 [论文]
67. [Arxiv 22] 面向深度图模型的忠实解释 [论文]
68. [Arxiv 22] 迈向无监督图级别表示学习的解释 [论文]
69. [Arxiv 22] BAGEL：评估图神经网络解释的基准 [论文]
70. [Arxiv 22] 图神经网络中的可解释性：一项实验性调查 [论文]
71. [IEEE TSIPN 22] 社交互动中的可解释性与图学习 [论文]

2021年

1. [NeurIPS 21] SALKG：基于知识图谱解释的常识推理学习 [论文]
2. [NeurIPS 2021] 用于图神经网络的强化学习增强型解释器 [论文]
3. [NeurIPS 2021] 迈向图神经网络的多粒度可解释性 [论文]
4. [NeurIPS 2021] 图神经网络上的鲁棒反事实解释 [论文]
5. [ICML 2021] 通过子图探索实现图神经网络可解释性研究[论文]
6. [ICML 2021] 图神经网络的生成式因果解释[论文]
7. [ICML 2021] 利用正交化与诱导稀疏性提升分子图神经网络可解释性[论文]
8. [ICML 2021] 带有超参数重要性解释的自动化图表示学习[论文]
9. [ICLR 21] 面向时序知识图谱预测的可解释子图推理 [论文]
10. [ICLR 2021] 使用可微边掩码解释自然语言处理中的图神经网络[论文]
11. [ICLR 2021] 用于子图识别的图信息瓶颈 [论文]
12. [KDD 2021] 当与真实标签比较不正确时：关于评估图神经网络解释方法的探讨[论文]
13. [KDD 2021] 用于脑网络可解释分类的反事实图 [论文]
14. [CVPR 2021] 计算病理学中图神经网络解释器的量化评估.[论文]
15. [NAACL 2021] 基于图网络的可解释病历诊断中反事实支持事实提取。[论文]
16. [AAAI 2021] 基于模体驱动的图表示对比学习 [论文]
17. [TPAMI 21] 通过相关游走实现图神经网络高阶解释 [论文]
18. [WWW 2021] 利用优化框架解释并统一图神经网络 [论文]
19. [Genome medicine 21] 解释图卷积神经网络的决策：乳腺癌转移预测中患者特异性分子子网络的作用 [论文]
20. [IJCKG 21] 知识图嵌入在电商应用中的作用：注意力机制推理、解释及可迁移规则 [论文]
21. [RuleML+RR 21] 结合亚符号与符号方法实现可解释性 [论文]
22. [PAKDD 21] SCARLET：基于可解释注意力的图神经网络，用于假新闻传播者预测 [论文]
23. [J. Chem. Inf. Model] 用可解释人工智能为分子着色，以评估其临床前相关性 [论文]
24. [BioRxiv 21] APRILE：利用可解释图神经网络探索药物副作用的分子机制 [论文]
25. [ISM 21] 通过扩展卷积神经网络的可解释性方法，实现图神经网络的边级解释 [论文]
26. [Arxiv 21] 借助信息流实现数字病理学中图神经网络的解释 [论文]
27. [Arxiv 21] SEEN：利用邻域解释来细化图神经网络的解释 [论文]
28. [Arxiv 21] 保留、促进还是攻击？通过拓扑扰动进行图神经网络解释 [论文]
29. [Arxiv 21] 面向可解释图神经网络的学习稀疏化 [论文]
30. [ICML workshop 21] GCExplainer：人机协同的概念驱动型图神经网络解释器 [论文]
31. [ICML workshop 21] 通过对抗训练实现可靠的图神经网络解释 [论文]
32. [ICML workshop 21] 借鉴表格数据思想重新构想图神经网络解释 [论文]
33. [ICML workshop 21] 迈向图神经网络解释的自动化评估 [论文]
34. [ICDM 2021] GNES：学习如何解释图神经网络 [论文]
35. [ICDM 2021] GCN-SE：将注意力机制作为动态图节点分类的可解释性工具 [论文]
36. [ICDM 2021] 图神经网络预测的多目标解释 [论文]
37. [CIKM 2021] 迈向自解释图神经网络 [论文]
38. [ECML PKDD 2021] GraphSVX：图神经网络的Shapley值解释 [论文]
39. [WiseML 2021] 基于可解释性的后门攻击对抗图神经网络 [论文]
40. [IJCNN 21] MEG：为深度图网络生成分子反事实解释 [论文]
41. [ICCSA 2021] 利用加权图神经网络解释器理解药物滥用社交网络 [论文]
42. [NeSy 21] 一种新的图神经网络解释概念 [论文]
43. [Information Fusion 21] 借助图神经网络实现多模态因果推断，推动可解释人工智能的信息融合 [论文]
44. [Patterns 21] hcga：用于网络表型分析的高度比较性图分析 [论文]

2020年及之前

1. [NeurIPS 2020] 面向图神经网络的参数化解释器。[论文]
2. [NeurIPS 2020] PGM-Explainer：用于图神经网络的概率图模型解释 [论文]
3. [KDD 2020] XGNN：迈向图神经网络的模型级解释 [论文]
4. [ACL 2020] GCAN：面向社交媒体上可解释假新闻检测的图感知协同注意力网络。论文
5. [Arxiv 2020] 包含稀疏可解释性的图神经网络 [论文]
6. [NeurIPS Workshop 20] 迈向可解释的消息传递网络：用于预测金属有机框架中二氧化碳吸附的模型 [论文]
7. [ICML工作坊2020] 对比图神经网络解释 [论文]
8. [ICML工作坊2020] 迈向数字病理学中的可解释图表示 [论文]
9. [NeurIPS研讨会2020] 利用分子反事实解释深度图网络 [论文]
10. [DataMod 2020] 探索基于图的神经网络用于自动脑肿瘤分割 [论文]
11. [OpenReview 20] 一种用于可微分发现图算法的框架 [论文]
12. [OpenReview 20] 因果筛选法用于解释图神经网络 [论文]
13. [Arxiv 20] 从图信号去噪视角理解图神经网络 [论文]
14. [Arxiv 20] 通过幂迭代理解图神经网络中的消息传递 [论文]
15. [IJCNN 20] GCN-LRP解释：探索图卷积网络的潜在注意力机制 [论文]
16. [CD-MAKE 20] 解释图神经网络以理解节点分类中的加权图特征 [论文]
17. [ICDM 19] 大规模图推理的可扩展解释[论文]

awesome-graph-explainability-papers 快速上手指南

awesome-graph-explainability-papers 并非一个可直接安装的软件库或框架，而是一个精选论文与资源列表，旨在汇总图神经网络（GNN）可解释性领域的综述、平台工具及核心研究论文。

本指南将指导开发者如何利用该列表获取最新研究成果，并快速搭建相关的实验环境（基于列表中推荐的主流平台）。

环境准备

由于该仓库主要提供文献索引和外部工具链接，使用前需准备以下基础环境以运行列表中推荐的代码库（如 PyTorch Geometric, DIG, GraphXAI 等）：

• 操作系统：Linux (推荐 Ubuntu 20.04+), macOS, 或 Windows (WSL2 推荐)
• Python 版本：3.8 - 3.11
• 核心依赖：
  • PyTorch (深度学习框架)
  • Git (克隆代码库)
• 硬件要求：建议配备 NVIDIA GPU 以加速 GNN 训练与解释生成过程（CPU 亦可运行小规模演示）

安装步骤

1. 获取资源列表

首先克隆该仓库到本地，以便查阅最新的论文列表和链接：

git clone https://github.com/your-target-repo/awesome-graph-explainability-papers.git
cd awesome-graph-explainability-papers

(注：请替换为实际的仓库地址，若仅为浏览，可直接访问 GitHub 网页版)

2. 搭建实验环境（推荐方案）

列表中 Platforms 章节推荐了多个主流工具库。以下是以最常用的 PyTorch Geometric (PyG) 为例的安装步骤，它是实现大多数 GNN 可解释性算法（如 GNNExplainer, PGExplainer）的基础。

步骤 A: 安装 PyTorch 访问 PyTorch 官网 获取适合你环境的命令。国内用户可使用清华源加速：

pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

(注：cu118 代表 CUDA 11.8，请根据实际显卡驱动版本调整，若无 GPU 可选用 cpu 版本)

步骤 B: 安装 PyTorch Geometric 使用官方源或国内镜像安装核心库及其依赖：

pip install pyg-lib torch-scatter torch-sparse torch-cluster torch-spline-conv -f https://data.pyg.org/whl/torch-2.0.0+cu118.html
pip install torch-geometric

步骤 C: 安装专用解释性工具库 (可选) 若需复现特定论文，可安装列表中提到的专用库，例如 GraphXAI：

git clone https://github.com/mims-harvard/graphxai.git
cd graphxai
pip install -e .

基本使用

本资源的核心用法是查阅论文并复用代码。以下展示如何基于 PyTorch Geometric 运行一个最简单的 GNN 可解释性示例（对应列表中 GNNExplainer 方法）。

1. 加载模型与数据

假设你已经有一个训练好的图分类模型 model 和一个图数据 data。

import torch
from torch_geometric.nn import GCNConv, global_mean_pool
from torch_geometric.datasets import MUTAG
from torch_geometric.loader import DataLoader

# 加载数据集
dataset = MUTAG(root='/tmp/MUTAG')
loader = DataLoader(dataset, batch_size=64, shuffle=True)

# 定义简单 GCN 模型
class GCN(torch.nn.Module):
    def __init__(self, num_node_features, num_classes):
        super().__init__()
        self.conv1 = GCNConv(num_node_features, 64)
        self.conv2 = GCNConv(64, num_classes)

    def forward(self, x, edge_index, batch):
        x = self.conv1(x, edge_index).relu()
        x = self.conv2(x, edge_index)
        return global_mean_pool(x, batch)

model = GCN(dataset.num_node_features, dataset.num_classes)
# 此处应添加模型训练代码...

2. 应用可解释性方法 (GNNExplainer)

使用 torch_geometric.explain 模块生成解释，找出对预测结果最重要的子图和节点特征。

from torch_geometric.explain import Explainer, GNNExplainer

# 初始化解释器
explainer = Explainer(
    model=model,
    algorithm=GNNExplainer(epochs=100),
    explanation_type='model',
    node_mask_type='attributes',
    edge_mask_type='object',
)

# 对第一个图数据进行解释
explanation = explainer(data[0].x, data[0].edge_index, target=0)

# 可视化或分析结果
print(f"重要节点特征掩码形状：{explanation.node_mask.shape}")
print(f"重要边掩码形状：{explanation.edge_mask.shape}")

# 若需可视化 (需安装 matplotlib)
# explanation.visualize_graph() 

3. 查阅前沿论文

在本地克隆的仓库目录中，打开 README.md 文件，根据 Surveys（综述）、Most Influential Papers（最具影响力论文）或 Year 2024/2025 分类，点击对应的 [paper] 链接阅读最新算法原理，并前往对应的 [Code] 链接获取复现代码。