Iterative Graph Neural Network Enhancement Using Explanations

Abstract

We formulate an XAI-based model improvement approach for Graph Neural Networks (GNN) for node classification, called Explanation Enhanced Graph Learning (EEGL). The goal is to improve predictive performance using explanations. EEGL is an iterative algorithm, which starts with a learned “vanilla” GNN and repeatedly uses frequent subgraph mining to find relevant patterns in explanation subgraphs, which are then analyzed further to obtain application-dependent features corresponding to the presence of certain subgraphs in the node neighborhoods. Giving an application-dependent algorithm for such an extension of the Weisfeiler-Leman (1-WL) algorithm has been posed as an open problem. We present the results of experiments on different synthetic datasets, compare them with other feature annotations, and analyse the training dynamics.