<ul><li>Graph anomaly detection is crucial for identifying unusual patterns in graph-based data with applications in web security and financial fraud detection.</li><li>Existing methods in graph anomaly detection rely on contrastive learning, which may be compromised by disruptive noise introduced by interfering edges.</li><li>To address this limitation, a Clean-View Enhanced Graph Anomaly Detection framework (CVGAD) is proposed, incorporating a multi-scale anomaly awareness module and a progressive purification module.</li><li>Experiments conducted on five benchmark datasets demonstrate the effectiveness of the CVGAD framework in improving anomaly detection performance.</li></ul>

Rethinking Contrastive Learning in Graph Anomaly Detection: A Clean-View Perspective

Discover more