X-AddGraph: Post-Hoc Explainability Framework for Recurrent Graph Anomaly Detection

A new research paper introduces X-AddGraph, a post-hoc explainability framework for AddGraph, a foundational GCN+GRU model for edge-level anomaly detection in dynamic graphs. The framework addresses the opacity of deep learning detectors, which provide scores but no reasons for flagged anomalies, a critical issue in regulated information systems requiring auditable decisions. X-AddGraph employs a Dual Spatial-Temporal Attribution (DSTA) mechanism with three components aligned to AddGraph's architectural modules: gradient-based relevance attribution over the adjacency structure (spatial), direct reading of contextual attention weights (short-term temporal), and a third component (long-term temporal) not detailed in the abstract. The paper is available on arXiv under identifier 2608.12441, submitted as a cross-announcement. This work is significant for the field of explainable AI in dynamic graph anomaly detection, offering a method to make automated decisions more transparent and trustworthy.