Robust Graph Neural-Network-Based Encoder for Node and Edge Deep Anomaly Detection on Attributed Networks

The task of identifying anomalous users on attributed social networks requires the detection whose profile attributes and network structure significantly differ from those majority reference profiles. GNN-based models are well-suited for addressing challenge integrating node into learning process because they can efficiently incorporate demographic data, activity patterns, other relevant information. Aggregate operations, such as sum or mean pooling, utilized by Graph Neural Networks (GNNs) to combine representations neighboring nodes within a graph. However, these aggregate operations cause problems in detecting nodes. There two main issues consider when utilizing GNNs. Firstly, presence may affect representation normal nodes, leading false positives. Secondly, be overlooked their is flattened during operation, negatives. proposed approach, AnomEn, robust graph neural developed anomaly detection. It addresses challenges positives negatives using weighted mechanism. This mechanism designed differentiate between node’s own features its neighbors placing greater emphasis less neighbors’ features. system preserve original characteristics, whether anomalous. work proposes not only network, namely, but also specific structures edges. AnomEn method serves encoder edge architectures was tested multiple datasets. Experiments were conducted validate effectiveness encoder. findings demonstrated robustness anomalies. outperforms existing methods tasks 5.63% 7.87%.