DDGAD: Diffusion-Based Graph Anomaly Detection Using Trajectory Dynamics
A recent study published on arXiv (2605.26446) presents DDGAD, a novel framework utilizing diffusion for detecting anomalies in graphs. This approach leverages trajectory dynamics to differentiate between standard and anomalous nodes, effectively tackling the issue of contamination spread in GCN-based methods. In this framework, normal nodes maintain consistent trajectories through diffusion regularization and reliability-aware neighborhood consensus, whereas anomalous nodes display erratic dynamics. Potential applications of this research encompass financial risk management, analysis of social networks, and cybersecurity measures.
Key facts
- DDGAD is a diffusion-based graph anomaly detection framework.
- It leverages trajectory dynamics to distinguish normal and anomalous nodes.
- Normal nodes exhibit consistent and stable representation trajectories.
- Anomalous nodes exhibit unstable and conflicting dynamics.
- The method addresses contamination propagation in GCN-based methods.
- Applications include financial risk control, social network analysis, and cybersecurity.
- The paper is available on arXiv with ID 2605.26446.
- The approach uses diffusion regularization and reliability-aware neighborhood consensus.
Entities
Institutions
- arXiv