XGRADATE: Graph Anomaly Detection via Multi-Scale Cross-Layer Contrastive Learning Networks

Graph anomaly detection is a crucial area in data mining and machine learning, with applications in network security, fraud detection, and health monitoring. Current methods, however, face limitations such as over-reliance on local node information or inefficiency with large-scale graphs. To address these issues, we propose XGRADATE, a novel graph anomaly detection network based on cross-layer contrastive multi-scale contrastive learning. XGRADATE comprises two components: the cross-layer contrastive and the graph contrastive network. The cross-layer contrastive module captures local features and structural information more effectively by considering embedding features from different layers of the graph convolutional network. The graph contrastive network module integrates various types of anomaly information to thoroughly consider global and local features. Our method is evaluated on multiple datasets, demonstrating its effectiveness and practicality in improving the accuracy of graph anomaly detection tasks. XGRADATE provides a new solution for anomaly detection in graph data by integrating anomaly information at different levels and capturing local features and structural information more effectively.