RagGAD: Rationale-Aware Conditional Gaussian Mixture Normalizing Flow for Unsupervised Graph Anomaly Detection

2026-08-16

AI Summary

This study addresses the limitations of homophily assumptions in graph anomaly detection, which often induce spurious correlations and fail to capture the behavioral diversity of normal nodes. We propose an unsupervised framework based on rationale-aware conditional Gaussian mixture normalizing flows. This approach employs an adaptive rationale decoupler to disentangle stable from spurious associations, integrating rationale/non-rationale Gaussian mixtures with a robust-fragile hybrid learning strategy to precisely model complex heterogeneous distributions within a structure-aware space. Experimental results demonstrate that the proposed framework effectively overcomes false homophily constraints, significantly outperforming state-of-the-art methods across multiple benchmark datasets. Consequently, this work substantially enhances detection robustness in complex scenarios by mitigating the adverse effects of misleading structural correlations.

Abstract

Graph anomaly detection aims to identify nodes that deviate from normal behavioral patterns within graphs. However, existing methods largely rely on the homophily assumption, which makes it difficult to distinguish spurious affinities and to capture the diverse behaviors of normal nodes,limiting their robustness in complex real-world scenarios. To address this problem, we propose RagGAD, an unsupervised graph anomaly detection framework based on rationale-aware conditional Gaussian mixture normalizing flow. RagGAD introduces an adaptive rationale disentangler to disentangle stable rationales from spurious correlations within node interrelationships, and further decomposes stable rationales into robust and fragile components. The learned rationales capture underlying interaction patterns that characterize normal behaviors under varying conditions, while anomalies emerge as deviations associated with unstable or spurious correlations. To model the intricate distributions of normal and abnormal nodes, RagGAD integrates rationale-non-rationale Gaussian mixture modeling with a robust-fragile rationale mixture learning strategy. By mitigating spurious homophilic correlations and embracing the heterogeneity of normal patterns, RagGAD identifies anomalies as low-density regions within a structure-aware distribution space. Extensive experiments on multiple benchmark datasets demonstrate that RagGAD outperforms state-of-the-art methods.

Problem

• Graph Anomaly Detection
• Homophily Assumption
• Spurious Correlations
• Unsupervised Learning

Innovation

• Rationale-Aware Disentanglement
• Conditional Gaussian Mixture Normalizing Flow
• Unsupervised Graph Anomaly Detection
• Spurious Correlation Mitigation
• Robust-Fragile Rationale Decomposition