Bi-directional Curriculum Learning for Graph Anomaly Detection: Dual Focus on Homogeneity and Heterogeneity

TL;DR

Graph Anomaly Detection (GAD) faces node heterogeneity that many curriculum-learning approaches fail to exploit. The authors propose Bi-directional Curriculum Learning (BCL), which trains two parallel GAD models focusing on homogeneity and heterogeneity, guided by a bi-directional difficulty measure computed from node representations $h_i$ via a two-layer GCN, and controlled by a continuous pacing function in the form of $g(t)$. A fusion step combines the two directional scores with $Score_{final}=\alpha Score_{homo}+(1-\alpha) Score_{hete}$, yielding robust anomaly rankings. Across seven datasets and ten detectors, BCL consistently improves performance, with linear pacing proving particularly effective, and ablation studies confirming the benefit of jointly leveraging both homogeneity and heterogeneity. The work provides a practical plug-in strategy for enhancing GAD in real-world graphs, along with guidance on scheduling and parameter choices.

Abstract

Graph anomaly detection (GAD) aims to identify nodes from a graph that are significantly different from normal patterns. Most previous studies are model-driven, focusing on enhancing the detection effect by improving the model structure. However, these approaches often treat all nodes equally, neglecting the different contributions of various nodes to the training. Therefore, we introduce graph curriculum learning as a simple and effective plug-and-play module to optimize GAD methods. The existing graph curriculum learning mainly focuses on the homogeneity of graphs and treats nodes with high homogeneity as easy nodes. In fact, GAD models can handle not only graph homogeneity but also heterogeneity, which leads to the unsuitability of these existing methods. To address this problem, we propose an innovative Bi-directional Curriculum Learning strategy (BCL), which considers nodes with higher and lower similarity to neighbor nodes as simple nodes in the direction of focusing on homogeneity and focusing on heterogeneity, respectively, and prioritizes their training. Extensive experiments show that BCL can be quickly integrated into existing detection processes and significantly improves the performance of ten GAD anomaly detection models on seven commonly used datasets.

Figures (5)

• Figure 1: The Impact of Training Scheduler
• Figure 2: The overall framework of BCL
• Figure 3: Comparison of Difficult Measure for CLNode and BCL
• Figure 4: The Impact of different $\alpha$ on the performance
• Figure 5: The Impact of Training Scheduler