Neighborhood Enhancement based Residual Analysis Model for Anomaly Detection on Attributed Graph

Attributed graph node anomaly detection is an unsupervised learning task aimed at automatically identifying anomalous nodes that significantly deviate from the majority of nodes in the graph. Residual analysis-based methods have garnered substantial attention due to their ability to detect various types of anomalies. However, existing residual analysis methods only consider attribute information when reconstructing node attributes, thereby disregarding the structural information contained in the adjacency matrix. Accordingly, this paper proposes the Neighborhood Enhanced Residual Analysis Model for Anomaly Detection (NERadar), which incorporates a distinction between self-representation coefficients of neighbor and non-neighbor nodes to improve anomaly detection. Experimental results demonstrate that NERadar achieve higher average detection accuracy compared to other comparison methods on three real datasets and three synthetic anomaly datasets, particularly for large-scale datasets, while also achieving significantly lower time overhead than the baseline method Radar.