<ul><li>This paper introduces a novel approach to generate subgraph explanations for graph neural networks (GNNs) that optimize multiple measures for explainability.</li><li>Existing GNN explanation methods typically focus on a single explainability measure, leading to biased explanations. The proposed approach, called skyline explanation, aims to simultaneously optimize multiple explainability measures.</li><li>The paper formulates skyline explanation generation as a multi-objective optimization problem and presents efficient algorithms based on an onion-peeling approach to improve explanations and provide quality guarantees.</li><li>Empirical verification using real-world graphs confirms the effectiveness, efficiency, and scalability of the proposed algorithms for generating subgraph explanations in GNNs.</li></ul>

Generating Skyline Explanations for Graph Neural Networks

Discover more