Anomaly Detection in Dynamic Graphs via Transformer

Detecting anomalies for dynamic graphs has drawn increasing attention due to their wide applications in social networks, e-commerce, and cybersecurity. Recent deep learning-based approaches have shown promising results over shallow methods. However, they fail address two core challenges of anomaly detection graphs: the lack informative encoding unattributed nodes difficulty learning discriminate knowledge from coupled spatial-temporal graphs. To overcome these challenges, this paper, we present a novel Transformer-based Anomaly Detection framework DYnamic (TADDY). Our constructs comprehensive node strategy better represent each node's structural temporal roles an evolving stream. Meanwhile, TADDY captures representation with patterns via graph transformer model. The extensive experimental demonstrate that our proposed outperforms state-of-the-art methods by large margin on six real-world datasets.