Research map / Graph learning / Graph self-supervised learning
Truncated Affinity Maximization: One-class Homophily Modeling for Graph Anomaly Detection
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use TAM instead of the strongest competing detector | GAD models · Amazon | AUROC: 0.5996 -> 0.7064 | Helped |
| Use TAM instead of the strongest competing detector | GAD models · Amazon | AUPRC: 0.1424 -> 0.2634 | Helped |
| Use TAM instead of the strongest competing detector | GAD models · Reddit | AUROC: 0.6028 -> 0.6023 | No clear effect |
| Use TAM instead of the strongest competing detector | GAD models · Reddit | AUPRC: 0.0449 -> 0.0446 | No clear effect |
| Truncate the graph using NSGT rather than retain the original graph (ablation) | TAM on the six datasets · BlogCatalog, ACM, Amazon, Facebook, Reddit, YelpChi | AUROC: RG: 0.6728, 0.7511, 0.4763, 0.8186, 0.5575, 0.4943; NSGT: 0.8235, 0.8830, 0.7120, 0.9105, 0.5938, 0.5449 | Helped |
| Replace NSGT with random edge dropping (ablation) | TAM on the six datasets · BlogCatalog, ACM, Amazon, Facebook, Reddit, YelpChi | AUROC: ED: 0.5678, 0.7162, 0.4574, 0.8641, 0.5641, 0.5014 | Hurt |
| Replace NSGT with similarity-based removal of the least similar edges (ablation) | TAM on the six datasets · BlogCatalog, ACM, Amazon, Facebook, Reddit, YelpChi | AUROC: SC: 0.6650, 0.8668, 0.5856, 0.6951, 0.6007, 0.4910 | Hurt |
| Add LAMNet rather than score raw affinity after truncation (ablation) | TAM on the six datasets · BlogCatalog, ACM, Amazon, Facebook, Reddit, YelpChi | AUROC: RTA: 0.7497, 0.8043, 0.6256, 0.8161, 0.5783, 0.5118; LAMNet: 0.8248, 0.8878, 0.7064, 0.9144, 0.5923, 0.5643 | Helped |
| Replace the affinity-maximization objective with graph reconstruction loss (ablation) | TAM on the six datasets · BlogCatalog, ACM, Amazon, Facebook, Reddit, YelpChi | AUROC: DOM: 0.7642, 0.8679, 0.5169, 0.7793, 0.5863, 0.5154; LAMNet: 0.8248, 0.8878, 0.7064, 0.9144, 0.5923, 0.5643 | Mixed |
| Use shared rather than independent GNN weights across LAMNets (ablation) | TAM · Six GAD datasets | AUROC: TAM-S: 0.8170, 0.8826, 0.7044, 0.9165, 0.6008, 0.5407; TAM: 0.8248, 0.8878, 0.7064, 0.9144, 0.6023, 0.5643 | Mixed |
| Add regularization to the affinity objective (ablation) | TAM · BlogCatalog and Facebook | AUROC and AUPRC | Helped |
| Increase regularization strength (ablation) | TAM on real-world datasets · Amazon, Reddit, YelpChi | AUROC and AUPRC | Hurt |
Closest papers by meaning
- ARC: A Generalist Graph Anomaly Detector with In-Context Learning (NeurIPS 2024 · similarity 0.90)
- Graph Anomaly Detection as Dynamical Transport: Training-Free Scoring via Empirical Bayes (NeurIPS 2026 · similarity 0.90)
- ADA-GAD: Anomaly-Denoised Autoencoders for Graph Anomaly Detection (AAAI 2024 · similarity 0.90)
- Generative Semi-supervised Graph Anomaly Detection (NeurIPS 2024 · similarity 0.90)
- Semi-supervised Graph Anomaly Detection via Robust Homophily Learning (NeurIPS 2025 · similarity 0.89)
- Open-Set Graph Anomaly Detection via Normal Structure Regularisation (ICLR 2025 · similarity 0.89)
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.