Graph Outlier Detection

Graph outlier detection aims to identify nodes in a network that deviate significantly from the norm, either structurally or in their attributes. Current research focuses on improving the accuracy and efficiency of detection methods, particularly addressing class imbalance issues in supervised learning and data leakage problems in unsupervised approaches. This involves exploring various model architectures, including graph neural networks and traditional methods like tree ensembles, and developing comprehensive benchmark tools for fair comparison and evaluation. The field's advancements are crucial for numerous applications, ranging from fraud detection and anomaly identification in complex systems to improving the reliability of large language models.