Cross-Domain Graph Anomaly Detection via Anomaly-Aware Contrastive Alignment

Cross-domain graph anomaly detection (CD-GAD) describes the problem of detecting anomalous nodes in an unlabelled target using auxiliary, related source graphs with labelled and normal nodes. Although it presents a promising approach to address notoriously high false positive issue detection, little work has been done this line research. There are numerous domain adaptation methods literature, but is difficult adapt them for GAD due unknown distributions anomalies complex node relations embedded data. To end, we introduce novel approach, namely Anomaly-aware Contrastive alignmenT (ACT), GAD. ACT designed jointly optimise: (i) unsupervised contrastive learning representations graph, (ii) anomaly-aware one-class alignment that aligns these while enforcing significant deviation from graph. In doing so, effectively transfers anomaly-informed knowledge learn class on without any specification distributions. Extensive experiments eight CD-GAD settings demonstrate our achieves substantially improved performance over 10 state-of-the-art methods. Code available at https://github.com/QZ-WANG/ACT.