Sound Logical Explanations for Mean Aggregation Graph Neural Networks
This work addresses the problem of interpretability for mean-aggregation GNNs in knowledge graph completion, offering a novel theoretical framework and practical tools for explainability, though it is incremental as it builds on existing explainability methods for GNNs.
The paper tackles the lack of explainability and expressivity results for graph neural networks using mean aggregation, by proving the precise class of monotonic rules that are sound for such networks with non-negative weights and providing a restricted fragment of first-order logic for explanations. Experiments show that restricting to non-negative weights yields comparable or improved performance on benchmarks, sound rules are obtained in practice, and these rules can generate insightful explanations and expose issues in trained models.
Graph neural networks (GNNs) are frequently used for knowledge graph completion. Their black-box nature has motivated work that uses sound logical rules to explain predictions and characterise their expressivity. However, despite the prevalence of GNNs that use mean as an aggregation function, explainability and expressivity results are lacking for them. We consider GNNs with mean aggregation and non-negative weights (MAGNNs), proving the precise class of monotonic rules that can be sound for them, as well as providing a restricted fragment of first-order logic to explain any MAGNN prediction. Our experiments show that restricting mean-aggregation GNNs to have non-negative weights yields comparable or improved performance on standard inductive benchmarks, that sound rules are obtained in practice, that insightful explanations can be generated in practice, and that the sound rules can expose issues in the trained models.