NK-GAD: Neighbor Knowledge-Enhanced Unsupervised Graph Anomaly Detection

TL;DR

NK-GAD introduces a neighbor knowledge-enhanced framework for unsupervised graph anomaly detection that effectively handles attribute heterophily, outperforming existing methods across multiple datasets.

Contribution

The paper proposes NK-GAD, a novel framework that integrates neighbor similarity and dissimilarity, with modules for reconstruction and feature refinement, addressing limitations of homophily-based approaches.

Findings

NK-GAD achieves an average 3.29% AUC improvement over baselines.

Analysis reveals attribute similarities are similar across connected node pairs regardless of anomaly status.

Spectral energy distribution trends differ between normal and anomalous edges, aiding detection.

Abstract

Graph anomaly detection aims to identify irregular patterns in graph-structured data. Most unsupervised GNN-based methods rely on the homophily assumption that connected nodes share similar attributes. However, real-world graphs often exhibit attribute-level heterophily, where connected nodes have dissimilar attributes. Our analysis of attribute-level heterophily graphs reveals two phenomena indicating that current approaches are not practical for unsupervised graph anomaly detection: 1) attribute similarities between connected nodes show nearly identical distributions across different connected node pair types, and 2) anomalies cause consistent variation trends between the graph with and without anomalous edges in the low- and high-frequency components of the spectral energy distributions, while the mid-part exhibits more erratic variations. Based on these observations, we propose NK-GAD, a neighbor knowledge-enhanced unsupervised graph anomaly detection framework. NK-GAD integrates a joint encoder capturing both similar and dissimilar neighbor features, a neighbor reconstruction module modeling normal distributions, a center aggregation module refining node features, and dual decoders for reconstructing attributes and structures. Experiments on seven datasets show NK-GAD achieves an average 3.29\% AUC improvement.