GRExplainer: A Universal Explanation Method for Temporal Graph Neural Networks

TL;DR

GRExplainer introduces a universal, efficient, and user-friendly approach to explain Temporal Graph Neural Networks by converting dynamic graphs into unified node sequences and employing an RNN-based generator to produce connected explanation subgraphs. It unifies input formats across snapshot- and event-based TGNNs through BFS-driven sequences and temporal ordering, achieving superior fidelity and cohesiveness while reducing computational costs to $O(MN)$ compared to edge-centric baselines. Across six real-world datasets and three TGNNs, GRExplainer yields substantial gains in fidelity (FID+) and sparsity-based metrics (AUFSC), with notable efficiency improvements (up to 16x faster in some cases) and better connectivity of explanations, all without requiring manual parameter selection. This work advances trustworthy AI for dynamic networks by enabling scalable, instance-level explanations that are coherent, generalizable, and readily applicable to real-world TGNN deployments.

Abstract

Dynamic graphs are widely used to represent evolving real-world networks. Temporal Graph Neural Networks (TGNNs) have emerged as a powerful tool for processing such graphs, but the lack of transparency and explainability limits their practical adoption. Research on TGNN explainability is still in its early stages and faces several key issues: (i) Current methods are tailored to specific TGNN types, restricting generality. (ii) They suffer from high computational costs, making them unsuitable for large-scale networks. (iii) They often overlook the structural connectivity of explanations and require prior knowledge, reducing user-friendliness. To address these issues, we propose GRExplainer, the first universal, efficient, and user-friendly explanation method for TGNNs. GRExplainer extracts node sequences as a unified feature representation, making it independent of specific input formats and thus applicable to both snapshot-based and event-based TGNNs (the major types of TGNNs). By utilizing breadth-first search and temporal information to construct input node sequences, GRExplainer reduces redundant computation and improves efficiency. To enhance user-friendliness, we design a generative model based on Recurrent Neural Networks (RNNs), enabling automated and continuous explanation generation. Experiments on six real-world datasets with three target TGNNs show that GRExplainer outperforms existing baseline methods in generality, efficiency, and user-friendliness.

Figures (9)

• Figure 1: GRExplainer overview.
• Figure 2: Generation model architecture.
• Figure 3: Fidelity-Sparsity comparison of GRExplainer and baselines.
• Figure 4: Fidelity-Sparsity comparison with and without temporal information.
• Figure 5: Efficiency comparison of GRExplainer and baselines.