GAD-NR: Graph Anomaly Detection via Neighborhood Reconstruction

2023-06-02

Web Search and Data Mining

AI Summary

Existing graph autoencoders (GAEs) exhibit limited performance in detecting non-clustered, structurally complex anomalous nodes. To address this, we propose the Neighborhood Reconstruction Graph Autoencoder (NR-GAE), the first GAE framework to introduce neighborhood-level (rather than edge-level) reconstruction for graph anomaly detection. NR-GAE jointly reconstructs the local topological structure, node attributes, and neighbor attributes of a target node, thereby unifying the modeling of three fundamental structural anomaly types. It employs a multi-granularity neighborhood reconstruction loss to guide graph neural networks in learning robust node representations. Extensive experiments on six real-world datasets demonstrate that NR-GAE achieves up to a 30% improvement in AUC over state-of-the-art methods. Moreover, it maintains consistently high detection robustness across diverse anomaly patterns—significantly overcoming the longstanding limitation of conventional GAEs, which are largely confined to clustered anomalies.

Abstract

Graph Anomaly Detection (GAD) is a technique used to identify abnormal nodes within graphs, finding applications in network security, fraud detection, social media spam detection, and various other domains. A common method for GAD is Graph Auto-Encoders (GAEs), which encode graph data into node representations and identify anomalies by assessing the reconstruction quality of the graphs based on these representations. However, existing GAE models are primarily optimized for direct link reconstruction, resulting in nodes connected in the graph being clustered in the latent space. As a result, they excel at detecting cluster-type structural anomalies but struggle with more complex structural anomalies that do not conform to clusters. To address this limitation, we propose a novel solution called proj, a new variant of GAE that incorporates neighborhood reconstruction for graph anomaly detection. proj aims to reconstruct the entire neighborhood of a node, encompassing the local structure, self-attributes, and neighbor attributes, based on the corresponding node representation. By comparing the neighborhood reconstruction loss between anomalous nodes and normal nodes, proj can effectively detect any anomalies. Extensive experimentation conducted on six real-world datasets validates the effectiveness of proj, showcasing significant improvements (by up to 30%uparrow in AUC) over state-of-the-art competitors. The hrefhttps://github.com/anonymoususer437/GAD-NR extcolorcyan source code for proj is openly available. Importantly, the comparative analysis reveals that the existing methods perform well only in detecting one or two types of anomalies out of the three types studied. In contrast, GAD-NR excels at detecting all three types of anomalies across the datasets, demonstrating its comprehensive anomaly detection capabilities.

Problem

• Graph Anomaly Detection
• Complex Irregularities
• GAEs Limitations

Innovation

• Graph Anomaly Detection
• Neighbor Reconstruction
• Comprehensive Detection