ST-TGExplainer: Disentangling Stability and Transition Patterns for Temporal GNN Interpretability

Researchers introduced ST-TGExplainer, a self-explainable temporal GNN for more faithful graph explanations.

The method separates stability patterns from transition patterns to capture both repeated and first-time interactions.

It uses a disentangled information bottleneck objective to learn compact, predictive explanatory subgraphs.

This addresses a major gap in temporal GNN interpretability and improves explanation faithfulness.