GRASPED: Graph Anomaly Detection using Autoencoder with Spectral Encoder and Decoder (Full Version)

TL;DR

GRASPED tackles node-level graph anomaly detection by combining a multiscale Graph Wavelet Neural NetworkEncoder with a Wiener-graph DeconvolutionalDecoder. The encoder captures spectral information across multiple resolutions, while the decoder reconstructs node features and neighborhood distributions using a learnable, diffusion-based deconvolution and a GDN-based attribute reconstruction, respectively. The model optimizes a joint reconstruction loss over node degree, neighbor distributions, and attributes, enabling unsupervised detection of anomalies via reconstruction errors. Experiments on five real-world graphs show robust, state-of-the-art performance and strong ablation and hyperparameter analyses, highlighting GRASPED’s ability to leverage spectral information and multi-scale graph structure for reliable anomaly detection with limited labeled data.

Abstract

Graph machine learning has been widely explored in various domains, such as community detection, transaction analysis, and recommendation systems. In these applications, anomaly detection plays an important role. Recently, studies have shown that anomalies on graphs induce spectral shifts. Some supervised methods have improved the utilization of such spectral domain information. However, they remain limited by the scarcity of labeled data due to the nature of anomalies. On the other hand, existing unsupervised learning approaches predominantly rely on spatial information or only employ low-pass filters, thereby losing the capacity for multi-band analysis. In this paper, we propose Graph Autoencoder with Spectral Encoder and Spectral Decoder (GRASPED) for node anomaly detection. Our unsupervised learning model features an encoder based on Graph Wavelet Convolution, along with structural and attribute decoders. The Graph Wavelet Convolution-based encoder, combined with a Wiener Graph Deconvolution-based decoder, exhibits bandpass filter characteristics that capture global and local graph information at multiple scales. This design allows for a learning-based reconstruction of node attributes, effectively capturing anomaly information. Extensive experiments on several real-world graph anomaly detection datasets demonstrate that GRASPED outperforms current state-of-the-art models.

Figures (5)

• Figure 1: Illustration of GRASPED
• Figure 2: Experiment results of the hyperparameter $K$ for all datasets.
• Figure 3: Experiment results of the hyperparameter $\beta$ for all datasets.
• Figure 4: Experiment results of the hyperparameter $S$ for all datasets.
• Figure 5: Results on the Cora dataset with different anomaly types and injection rates.