GraphXAI: Synthetic and Real-World Graph Datasets with Ground-Truth Explanations

GraphXAI is an open-source library providing a synthetic graph data generator and XAI-ready real-world datasets for evaluating Graph Neural Network explainers. The library includes ShapeGGen, which can generate benchmark datasets with varying graph sizes, degree distributions, and homophily levels, all accompanied by ground-truth explanations. It was created by Owen Queen of Harvard University Press and is available on paperswithcode.

Use Cases

• Benchmarking GNN explanation methods based on the provided synthetic datasets with ground-truth explanations.
• Evaluating explanation reliability on heterophilic versus homophilic graph structures generated by ShapeGGen.
• Testing GNN explainers on real-world graph datasets that have been processed to include ground-truth explanations.
• Mimicking data from real-world applications using the flexible synthetic data generator to create diverse test scenarios.

Strengths

• Includes a synthetic data generator (ShapeGGen) capable of producing datasets with controlled properties like graph size and homophily level.
• Provides ground-truth explanations for generated and included real-world datasets, which are often unavailable.
• Offers a broader ecosystem including data loaders, processing functions, visualizers, GNN model implementations, and evaluation metrics.

Limitations

• Row count and dataset size are unknown, which may limit suitability assessment for large-scale experiments.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

paperswithcode

Collection Method

Includes synthetically generated data via the ShapeGGen tool and curated real-world graph datasets.