Research map / Graph learning / Graph self-supervised learning
Counterfactual Graph Learning for Anomaly Detection with Feature Disentanglement and Generation (Student Abstract)
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes CFGAD, a counterfactual graph anomaly detector that uses gradient-based feature separation into class and environment features and a weight-varying diffusion model to generate counterfactual node features for robust graph anomaly detection.
Problem
Graph anomaly detection models learn dataset-specific spurious correlations and degrade on newly observed nodes under different environments (distribution shift and shortcut learning).
Approach
Disentangle node features via gradient-based importance scores into class and environment features, then recombine them with a weight-varying conditional diffusion model to synthesize counterfactual samples that augment training for robustness.
Key results
- CFGAD reaches 0.6832 macro-F1 and 0.8129 AUC on YelpChi, versus 0.6577 / 0.7853 for the best baseline GraphConsis-level comparison (CARE-GNN 0.6433 / 0.7925).
- On Amazon, CFGAD achieves 0.9054 macro-F1 and 0.9629 AUC, outperforming CARE-GNN (0.8988 / 0.9491).
- The paper reports CFGAD outperforming all listed baselines (GCN, GAT, GraphConsis, CARE-GNN) on both datasets.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use CFGAD instead of GraphConsis | Graph anomaly detection · YelpChi | macro-F1: 0.6577 -> 0.6832 | Helped |
| Use CFGAD instead of CARE-GNN | Graph anomaly detection · YelpChi | AUC: 0.7925 -> 0.8129 | Helped |
| Use CFGAD instead of CARE-GNN | Graph anomaly detection · Amazon | macro-F1: 0.8988 -> 0.9054 | Helped |
| Use CFGAD instead of GraphConsis | Graph anomaly detection · Amazon | AUC: 0.9516 -> 0.9629 | Helped |
Limitations
- Evaluation limited to two datasets and four baselines in a short student abstract.
- No ablation or analysis of the diffusion weight parameters or top-K selection reported in the text.
- Detailed hyperparameters, training setup, and hardware not described.
Experimental setup
Datasets: YelpChi Amazon
Benchmarks: YelpChi Amazon
Metrics: macro-F1 AUC
Baselines: GCN GAT GraphConsis CARE-GNN
Builds on: CARE-GNN Diffusion models
Closest papers by meaning
- DiffGAD: A Diffusion-based Unsupervised Graph Anomaly Detector (ICLR 2025 · similarity 0.89)
- Leveraging Diffusion Model as Pseudo-Anomalous Graph Generator for Graph-Level Anomaly Detection (ICML 2025 · similarity 0.86)
- ARC: A Generalist Graph Anomaly Detector with In-Context Learning (NeurIPS 2024 · similarity 0.85)
- UniFORM: Towards Unified Framework for Anomaly Detection on Graphs (AAAI 2025 · similarity 0.85)
- Unifying Unsupervised Graph-Level Anomaly Detection and Out-of-Distribution Detection: A Benchmark (ICLR 2025 · similarity 0.85)
- UniGAD: Unifying Multi-level Graph Anomaly Detection (NeurIPS 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.