Research map / Graph learning / Graph self-supervised learning
Graph Anomaly Detection with Diffusion Model-Based Graph Enhancement (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 DEGAD, a contrastive graph anomaly detection framework that uses a diffusion-based graph enhancement module to manipulate neighbors and reduce inconsistent positive pairs, with a multi-scale node-subgraph contrastive module.
Problem
Contrastive graph anomaly detection forces node-level and subgraph-level representations with different class labels (common under heterophily) to agree, which hurts robustness and detection quality.
Approach
Use a diffusion model to inject features of reference nodes into a portion of the target node's neighbors, producing enhanced graphs whose subgraph representations better match the node's label; then apply a multi-scale node-subgraph contrastive loss over two enhanced graphs.
Key results
- DEGAD reaches F1 0.7101 / AUC 0.9302 on Cora, outperforming GRADATE (0.7066 / 0.9237).
- On CiteSeer, DEGAD achieves F1 0.8241 and AUC 0.9504, versus GRADATE's 0.8092 and 0.9409.
- On PubMed, DEGAD achieves F1 0.9153 and AUC 0.9852, versus GRADATE's 0.9005 and 0.9820.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use DEGAD instead of GRADATE | Graph anomaly detection · Cora | F1: 0.7066 -> 0.7101 | Helped |
| Use DEGAD instead of GRADATE | Graph anomaly detection · Cora | AUC: 0.9237 -> 0.9302 | Helped |
| Use DEGAD instead of GRADATE | Graph anomaly detection · CiteSeer | F1: 0.8092 -> 0.8241 | Helped |
| Use DEGAD instead of GRADATE | Graph anomaly detection · CiteSeer | AUC: 0.9409 -> 0.9504 | Helped |
| Use DEGAD instead of GRADATE | Graph anomaly detection · PubMed | F1: 0.9005 -> 0.9153 | Helped |
| Use DEGAD instead of GRADATE | Graph anomaly detection · PubMed | AUC: 0.9820 -> 0.9852 | Helped |
Limitations
- Evaluated only on three small citation-style benchmarks with injected anomalies (Cora, CiteSeer, PubMed).
- Gains over GRADATE are small on Cora and CiteSeer.
- Short student abstract; implementation details, hyperparameters, and compute are not reported.
Experimental setup
Datasets: Cora CiteSeer PubMed
Benchmarks: Cora CiteSeer PubMed
Metrics: F1 AUC
Baselines: Autoencoder DOMINANT ALARM CoLA ANEMONE GRADATE
Builds on: CoLA ANEMONE GRADATE DOMINANT denoising diffusion probabilistic model
Closest papers by meaning
- DiffGAD: A Diffusion-based Unsupervised Graph Anomaly Detector (ICLR 2025 · similarity 0.90)
- Leveraging Diffusion Model as Pseudo-Anomalous Graph Generator for Graph-Level Anomaly Detection (ICML 2025 · similarity 0.89)
- ADA-GAD: Anomaly-Denoised Autoencoders for Graph Anomaly Detection (AAAI 2024 · similarity 0.87)
- Revisiting Graph Contrastive Learning on Anomaly Detection: A Structural Imbalance Perspective (AAAI 2025 · similarity 0.87)
- Exploitation of a Latent Mechanism in Graph Contrastive Learning: Representation Scattering (NeurIPS 2024 · similarity 0.86)
- ReGCL: Rethinking Message Passing in Graph Contrastive Learning (AAAI 2024 · similarity 0.86)
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.