Structural-Temporal Coupling Anomaly Detection with Dynamic Graph Transformer

2025-05-13

AI Summary

To address the insufficient discriminative power of node-pair representations caused by decoupling structural and temporal features in dynamic graph anomaly edge detection, this paper introduces, for the first time, a modeling framework for structure–time coupled evolution patterns. We propose a dynamic graph Transformer architecture featuring dual-granularity feature fusion and two-dimensional positional encoding to jointly capture multi-scale structural and temporal dependencies. By synergistically integrating dynamic graph neural networks with Transformer-based sequence modeling, our method explicitly characterizes the co-evolution of node pairs across both topological and temporal dimensions. Evaluated on six benchmark datasets, the model achieves significant improvements over existing state-of-the-art methods. In real-world social and financial transaction scenarios, it demonstrates superior accuracy—yielding average AUC gains of 3.2–7.8%—and enhanced interpretability, enabling fine-grained anomaly attribution analysis.

Technology Category

• Machine Learning: Graph-based Machine Learning
• Data Mining & Knowledge Management: Graph Mining, Social Network Analysis & Community
• Reasoning under Uncertainty: Graphical Models

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Web Mining and Content Analysis: Bridging structured and unstructured data
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Detecting anomalous edges in dynamic graphs is an important task in many applications over evolving triple-based data, such as social networks, transaction management, and epidemiology. A major challenge with this task is the absence of structural-temporal coupling information, which decreases the ability of the representation to distinguish anomalies from normal instances. Existing methods focus on handling independent structural and temporal features with embedding models, which ignore the deep interaction between these two types of information. In this paper, we propose a structural-temporal coupling anomaly detection architecture with a dynamic graph transformer model. Specifically, we introduce structural and temporal features from two integration levels to provide anomaly-aware graph evolutionary patterns. Then, a dynamic graph transformer enhanced by two-dimensional positional encoding is implemented to capture both discrimination and contextual consistency signals. Extensive experiments on six datasets demonstrate that our method outperforms current state-of-the-art models. Finally, a case study illustrates the strength of our method when applied to a real-world task.

Problem

• Detecting anomalous edges in dynamic graphs
• Addressing absence of structural-temporal coupling information
• Capturing deep interaction between structural and temporal features

Innovation

• Dynamic graph transformer for anomaly detection
• Two-dimensional positional encoding enhancement
• Structural-temporal coupling integration levels