Measuring What Matters: A Unified Evaluation Framework for GNN Explainability

2026-07-05

AI Summary

This work addresses the lack of a unified, ground-truth-free evaluation framework for explainability methods in graph neural networks (GNNs), which hinders the reliable selection and practical deployment of interpreters. We propose the first evaluation framework that operates without assumptions about ground-truth explanations, adapting interpretability metrics from tabular data to the graph domain. By decoupling the contributions of graph topology and node features, our approach enables multi-dimensional quantitative assessment. Through large-scale benchmarking and Pareto front analysis, we identify non-dominated explainers that exhibit robust performance across diverse tasks, demonstrating that no single method universally outperforms others. To support real-world adoption, we further release a practitioner-oriented guide for deploying and evaluating GNN explainability (G-XAI) methods.

Abstract

Graph eXplainable AI (G-XAI) is increasingly important for making Graph Neural Networks interpretable and accountable. While a growing number of explainers are available, choosing the right method and assessing the trustworthiness of its outputs remains unclear. Consistent evaluation practices and actionable guidance are still missing, hindering practical adoption. In this paper, we introduce a unified, quantitative benchmarking framework for G-XAI that requires no ground-truth assumptions. We formalize tabular explainability metrics for graph data, evaluating topological structure and node features as independent components. Our large-scale benchmarking study identifies explainers that consistently lie on the Pareto front across metric pairs and tasks, establishing robustly non-dominated solutions - while confirming that no single explainer achieves universal superiority. We distill our findings into actionable G-XAI usability guidelines to support Machine Learning practitioners in evaluating and deploying trustworthy GNN-based pipelines.

Problem

• Graph eXplainable AI
• GNN explainability
• evaluation framework
• trustworthiness
• benchmarking

Innovation

• Graph eXplainable AI
• unified evaluation framework
• ground-truth-free benchmarking
• Pareto-optimal explainers
• interpretable GNNs