Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Structural-Temporal Coupling Anomaly Detection with Dynamic Graph Transformer

Chang Zong, Yueting Zhuang, Jian Shao, Weiming Lu

TL;DR

STCAD tackles anomaly detection in dynamic graphs by integrating structural-temporal coupling into a dynamic graph transformer with a novel 2D positional encoding. It introduces two-level anomaly-aware features and a mixed supervision framework that jointly captures discriminative and contextual signals. Across six benchmark datasets, STCAD consistently surpasses state-of-the-art baselines, with ablation studies highlighting the importance of coupling features and 2D encoding. A case study on emerging technology identification demonstrates the method's practical utility for discovering novel domain co-occurrences and evolving patterns.

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.

Structural-Temporal Coupling Anomaly Detection with Dynamic Graph Transformer

TL;DR

STCAD tackles anomaly detection in dynamic graphs by integrating structural-temporal coupling into a dynamic graph transformer with a novel 2D positional encoding. It introduces two-level anomaly-aware features and a mixed supervision framework that jointly captures discriminative and contextual signals. Across six benchmark datasets, STCAD consistently surpasses state-of-the-art baselines, with ablation studies highlighting the importance of coupling features and 2D encoding. A case study on emerging technology identification demonstrates the method's practical utility for discovering novel domain co-occurrences and evolving patterns.

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.
Paper Structure (32 sections, 16 equations, 7 figures, 2 tables, 1 algorithm)

This paper contains 32 sections, 16 equations, 7 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Dynamic Graph Anomaly Detection
  4. Transformer and Graph Transformer Models
  5. Problem Formulation
  6. Methodology
  7. Structural-Temporal Node Encoder
  8. Edges and Context Sampling:
  9. Two-Level Feature Encoding:
  10. Node Features Aggregation:
  11. Dynamic Graph Transformer
  12. Input Sequence Construction:
  13. 2D Positional Encoding:
  14. Transformer Edge Encoding:
  15. Mixed Supervision Anomaly Detector
  16. ...and 17 more sections

Figures (7)

  • Figure 1: Architecture of our proposed STCAD for anomalous edges detection in dynamic graphs.
  • Figure 2: Performance reported on six benchmark datasets with AUC and AP metrics. Three anomaly proportions are evaluated. The best result for each experiment is highlighted in bold.
  • Figure 3: Impact of three coupling features on AUC and AP metrics. Four datasets are evaluated including UCI-Message (top left), Digg (top right), AS-Topology (bottom left), and Bitcoin-Alpha (bottom right). Best results on each dataset are marked with inverted triangles.
  • Figure 4: Performance with different hyperparameters on three datasets. The best results are highlighted with bold boxes.
  • Figure 5: Latent space embeddings generated by STCAD on six datasets. Red dots are anomalous edges and blue dots are normal samples.
  • ...and 2 more figures