A Comprehensive Survey on Graph Anomaly Detection with Deep Learning

Anomalies represent rare observations (e.g., data records or events) that deviate significantly from others. Over several decades, research on anomaly mining has received increasing interests due to the implications of these occurrences in a wide range disciplines. Anomaly detection, which aims identify observations, is among most vital tasks world, and shown its power preventing detrimental events, such as financial fraud, network intrusion, social spam. The detection task typically solved by identifying outlying points feature space inherently overlooks relational information real-world data. Graphs have been prevalently used structural information, raises graph problem - anomalous objects (i.e., nodes, edges sub-graphs) single graph, graphs database/set graphs. However, conventional techniques cannot tackle this well because complexity For advent deep learning, with learning growing attention recently. In survey, we aim provide systematic comprehensive review contemporary for detection. We compile open-sourced implementations, public datasets, commonly-used evaluation metrics affluent resources future studies. More importantly, highlight twelve extensive directions according our survey results covering unsolved emerging problems applications. With goal create "one-stop-shop" provides unified understanding categories existing approaches, publicly available hands-on resources, high-impact open challenges using learning.