Research map / Label-efficient and robust learning / OOD and anomaly detection
Graph Anomaly Detection via Prototype-Aware Label Propagation (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 LP-GAD, a graph anomaly detection model that uses label propagation with structure and attribute-influence-based pseudo labels, refined by a prototype-aware strategy, and reports AUC gains over SSL and semi-supervised GAD baselines on six datasets.
Problem
Graph anomaly detection on attributed graphs where labelled anomalies are scarce, and anomalies may hide in structure, attribute, or both spaces; existing semi-supervised methods overfit with few labels and assume homophily.
Approach
Generates pseudo labels via a structure label propagation (Markov diffusion with learnable self-gated operator weights) and an attribute pseudo label from node-influence-reweighted attributes (degree times local density), then corrects and refines them with class prototypes in the target network embedding space using a Jaccard-weighted moving average.
Key results
- LP-GAD achieves AUC 82.95 on BlogCatalog and 98.14 on PubMed, the best in Table 1 except on Cora.
- Reports a 2.97% AUC improvement on Flickr over SSL baselines.
- Reports a 2.84% F1-macro improvement on PubMed over SSL baselines.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use LP-GAD instead of SL-GAD | Graph anomaly detection · BlogCatalog | AUC (%): 81.84 -> 82.95 | Helped |
| Use LP-GAD instead of PC-GNN | Graph anomaly detection · Flickr | AUC (%): 79.75 -> 82.62 | Helped |
| Use LP-GAD instead of SL-GAD | Graph anomaly detection · ACM | AUC (%): 85.38 -> 86.89 | Helped |
| Use LP-GAD instead of PC-GNN | Graph anomaly detection · Cora | AUC (%): 91.32 -> 90.76 | Hurt |
| Use LP-GAD instead of PC-GNN | Graph anomaly detection · Citeseer | AUC (%): 93.03 -> 93.80 | Helped |
| Use LP-GAD instead of PC-GNN | Graph anomaly detection · PubMed | AUC (%): 97.09 -> 98.14 | Helped |
Surprising findings
- LP-GAD does not achieve the best result on Cora, where it is outperformed by PC-GNN, SL-GAD, and ANEMONE on AUC.
Limitations
- Label propagation is said to favour homophilic graphs, which the Markov diffusion design only partly addresses.
- Results on Cora are not best, and the paper does not analyse why.
- Short student abstract; implementation and hardware details are sparse.
Experimental setup
Datasets: BlogCatalog Flickr ACM Cora Citeseer PubMed
Benchmarks: BlogCatalog Flickr ACM Cora Citeseer PubMed
Metrics: AUC F1-macro
Baselines: GAE DOMINANT ANEMONE CoLA GCCAD Meta-GDN SL-GAD PC-GNN
Builds on: label propagation SL-GAD PC-GNN
Closest papers by meaning
- ARC: A Generalist Graph Anomaly Detector with In-Context Learning (NeurIPS 2024 · similarity 0.87)
- DiffGAD: A Diffusion-based Unsupervised Graph Anomaly Detector (ICLR 2025 · similarity 0.87)
- Graph Anomaly Detection as Dynamical Transport: Training-Free Scoring via Empirical Bayes (NeurIPS 2026 · similarity 0.87)
- Open-Set Graph Anomaly Detection via Normal Structure Regularisation (ICLR 2025 · similarity 0.86)
- Leveraging Diffusion Model as Pseudo-Anomalous Graph Generator for Graph-Level Anomaly Detection (ICML 2025 · similarity 0.86)
- UniGAD: Unifying Multi-level Graph Anomaly Detection (NeurIPS 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.