A Masked AutoEncoder with Strong-Weak Mutual Information for Anomaly Detection in Dynamic Incomplete Graphs

Most graph anomaly detection methods leverage GNNs to learn from relatively high-quality graph data. Unfortunately, such ideal scenarios are rare in real-world applications, where most data suffer from issues such as missing labels, dynamic changes, and incompleteness, collectively referred to as Dynamic Incomplete Graphs (DIGs). To address the challenges of GNNs failing under extreme conditions, we propose the Graph Masked AutoEncoder with Strong-Weak Mutual Information (GMAE-SWMI). This framework simulates real-world DIG scenarios by masking graph structures (nodes/edges) and node features. Additionally, through the SWMI loss, it captures the relationship between structure and features while maintaining structural integrity, reducing overfitting, and improving generalization. We evaluated our approach on eight real-world graph datasets, demonstrating its superiority over state-of-the-art methods across different levels of DIG scenarios across various downstream tasks and representation evaluations, both with and without labels.