<ul><li>Graph neural networks (GNNs) are widely used in various domains such as social networks, molecular biology, or recommendation systems.</li><li>Explanations of GNNs' predictions can be factual or counterfactual, with counterfactual explanations involving transforming 'reject' graphs into 'accept' graphs.</li><li>A common recourse explanation generates a small set of 'accept' graphs relevant to all input 'reject' graphs, applicable for binary classification tasks.</li><li>Researchers introduce an algorithm, COMRECGC, to address the common recourse explanation problem for global counterfactual explanations in GNNs, demonstrating superior performance in real-world datasets.</li></ul>

COMRECGC: Global Graph Counterfactual Explainer through Common Recourse

Discover more