Research map / Graph learning / Graph self-supervised learning
GraphCleaner: Detecting Mislabelled Samples in Popular Graph Learning Benchmarks
Yuwen Li, Miao Xiong, Bryan Hooi
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 neighborhood-aware detection with synthetic mislabels generated from a learned class-transition matrix | GraphCleaner · Cora, 10% symmetric noise | F1: 0.560 -> 0.790 | Improved |
| Use neighborhood-aware detection with synthetic mislabels generated from a learned class-transition matrix | GraphCleaner · CiteSeer, 10% symmetric noise | F1: 0.432 -> 0.560 | Improved |
| Use neighborhood-aware detection with synthetic mislabels generated from a learned class-transition matrix | GraphCleaner · PubMed, 10% symmetric noise | F1: 0.447 -> 0.616 | Improved |
| Use neighborhood-aware detection with synthetic mislabels generated from a learned class-transition matrix | GraphCleaner · Computers, 10% symmetric noise | F1: 0.755 -> 0.844 | Improved |
| Use neighborhood-aware detection with synthetic mislabels generated from a learned class-transition matrix | GraphCleaner · Photo, 10% symmetric noise | F1: 0.777 -> 0.902 | Improved |
| Use neighborhood-aware detection with synthetic mislabels generated from a learned class-transition matrix | GraphCleaner · OGB-arxiv, 10% symmetric noise | F1: 0.500 -> 0.760 | Improved |
| Remove neighborhood-label agreement features (ablation) | GraphCleaner · Cora, Computers, and OGB-arxiv, 10% symmetric noise | F1: Cora: 0.798 -> 0.790; Computers: 0.840 -> 0.844; OGB-arxiv: 0.750 -> 0.760 | Mixed |
| Remove neighborhood-prediction agreement features (ablation) | GraphCleaner · Cora, Computers, and OGB-arxiv, 10% symmetric noise | F1: Cora: 0.676 -> 0.790; Computers: 0.844 -> 0.844; OGB-arxiv: 0.736 -> 0.760 | Mixed |
| Replace transition-matrix-guided synthetic noise with randomly selected incorrect labels (ablation) | GraphCleaner · 10% symmetric noise, Cora, Computers, and OGB-arxiv | F1: Cora: 0.802 -> 0.790; Computers: 0.865 -> 0.844; OGB-arxiv: 0.819 -> 0.760 | Mixed |
| Replace transition-matrix-guided synthetic noise with randomly selected incorrect labels (ablation) | GraphCleaner · 10% asymmetric noise, Cora, Computers, and OGB-arxiv | F1: Cora: 0.635 -> 0.669; Computers: 0.474 -> 0.778; OGB-arxiv: 0.295 -> 0.477 | Mixed |
| Remove noisy-labelled samples from the PubMed test set | GCN+mixup · PubMed test set | accuracy: 86.71% -> 89.11% | Improved |
| Use one-hop neighborhood information (ablation) | GraphCleaner · OGB-arxiv, 10% noise | F1 | No clear effect |
Closest papers by meaning
- No Metric to Rule Them All: Toward Principled Evaluations of Graph-Learning Datasets (ICML 2025 · similarity 0.87)
- ML-GOOD: Towards Multi-Label Graph Out-Of-Distribution Detection (AAAI 2025 · similarity 0.85)
- Position: Graph Learning Will Lose Relevance Due To Poor Benchmarks (ICML 2025 · similarity 0.85)
- DiffGAD: A Diffusion-based Unsupervised Graph Anomaly Detector (ICLR 2025 · similarity 0.85)
- NoisyGL: A Comprehensive Benchmark for Graph Neural Networks under Label Noise (NeurIPS 2024 · similarity 0.85)
- Graph Anomaly Detection via Prototype-Aware Label Propagation (Student Abstract) (AAAI 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.
