GECo: a community-based graph neural network explainer

Abstract

Graph Neural Networks (GNNs) are powerful models that can manage complex data sources and their interconnection links. Still, one of GNNs’ main drawbacks is their lack of explainability, which limits their application in sensitive fields. In this paper, we introduce a new approach involving graph communities to address the explainability of graph classification problems, called Graph Neural Network Explainer based on Communities (GECo). Taking into consideration that a community is a subset of graph nodes that are densely connected, GECo exploits the idea that these subgraphs should play a key role in graph classification. This assumption is reasonable, especially if we consider the message-passing mechanism, which is the basic mechanism of GNNs. GECo analyzes the contribution to the classification result of the communities in the graph, building a mask that highlights graph-relevant structures. GECo is tested for Graph Convolutional Networks on six artificial and four real-world graph datasets and is compared to the main explainability methods, such as PGMExplainer, PGExplainer, GNNExplainer, TAGE, and SubgraphX, using six different metrics. The obtained results outperform the other methods for artificial graph datasets and most real-world datasets. Furthermore, GECo is faster than most of the competitors since it does not need to make use of any sampling process.