Explaining Graph Neural Networks for Node Similarity on Graphs
This work addresses the need for explainable similarity search in applications like citation networks or knowledge graphs, but it is incremental as it evaluates existing explanation methods rather than introducing new ones.
The paper tackled the problem of explaining graph neural networks (GNNs) for node similarity search on graphs, finding that gradient-based explanations are actionable, consistent, and can be pruned for sparsity, unlike mutual information explanations.
Similarity search is a fundamental task for exploiting information in various applications dealing with graph data, such as citation networks or knowledge graphs. While this task has been intensively approached from heuristics to graph embeddings and graph neural networks (GNNs), providing explanations for similarity has received less attention. In this work we are concerned with explainable similarity search over graphs, by investigating how GNN-based methods for computing node similarities can be augmented with explanations. Specifically, we evaluate the performance of two prominent approaches towards explanations in GNNs, based on the concepts of mutual information (MI), and gradient-based explanations (GB). We discuss their suitability and empirically validate the properties of their explanations over different popular graph benchmarks. We find that unlike MI explanations, gradient-based explanations have three desirable properties. First, they are actionable: selecting inputs depending on them results in predictable changes in similarity scores. Second, they are consistent: the effect of selecting certain inputs overlaps very little with the effect of discarding them. Third, they can be pruned significantly to obtain sparse explanations that retain the effect on similarity scores.