mims-harvard/GraphXAI

Provides XAI-ready benchmark datasets with ground-truth explanations via the ShapeGGen generator—parameterizable for graph size, degree distribution, homophily, and fairness properties—alongside implementations of state-of-the-art explainers, evaluation metrics, and GNN models. Addresses the critical gap of reliable evaluation data for GNN explainability by enabling controlled generation of graphs with known subgraph explanations across varying structural and fairness scenarios.

206 stars. No commits in the last 6 months.