GLADformer: A Mixed Perspective for Graph-level Anomaly Detection

TL;DR

This work addresses graph-level anomaly detection (GLAD) by integrating global spectral reasoning with local multi-frequency information. It presents GLADformer, comprising a Spectrum-Enhanced Graph Transformer that leverages a super-node, RRWP-based positional encoding, and Rayleigh-quotient–driven spectral features, plus a Beta-band wavelet GNN for local spectral message passing, augmented by a variation-optimized cross-entropy loss $\mathcal{L}_{VOCE}$. The approach demonstrates state-of-the-art performance and robustness across ten real-world datasets, highlighting improved capture of global anomaly representations and spectral characteristics. By jointly modeling global spectral energy distributions and local band-pass information, GLADformer advances GLAD with practical impact for domains requiring reliable anomaly discrimination in graph-structured data.

Abstract

Graph-Level Anomaly Detection (GLAD) aims to distinguish anomalous graphs within a graph dataset. However, current methods are constrained by their receptive fields, struggling to learn global features within the graphs. Moreover, most contemporary methods are based on spatial domain and lack exploration of spectral characteristics. In this paper, we propose a multi-perspective hybrid graph-level anomaly detector namely GLADformer, consisting of two key modules. Specifically, we first design a Graph Transformer module with global spectrum enhancement, which ensures balanced and resilient parameter distributions by fusing global features and spectral distribution characteristics. Furthermore, to uncover local anomalous attributes, we customize a band-pass spectral GNN message passing module that further enhances the model's generalization capability. Through comprehensive experiments on ten real-world datasets from multiple domains, we validate the effectiveness and robustness of GLADformer. This demonstrates that GLADformer outperforms current state-of-the-art models in graph-level anomaly detection, particularly in effectively capturing global anomaly representations and spectral characteristics.

Figures (3)

• Figure 1: Examples for different categories of graph-level anomalies.
• Figure 2: The above image presents the overview of our model GLADformer, where (a) and (b) demonstrate the the spectrum-enhanced graph-transformer module and the local spectral message passing module respectively.
• Figure 3: Visualization analysis on NCI1 dataset.