Research map / Graph learning / Graph self-supervised learning
Hybrid-Collaborative Augmentation and Contrastive Sample Adaptive-Differential Awareness for Robust Attributed Graph Clustering
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 RAGC, a contrastive attributed graph clustering method that jointly augments node-level and edge-level embeddings (HCA) and reweights contrastive pairs by pseudo-label confidence and hard/easy and positive/negative status (CSADA), evaluated on six datasets.
Problem
Contrastive attributed graph clustering relies on node-level augmentation only and treats all contrastive sample pairs equally, limiting discriminability and robustness to noise.
Approach
Combine node-level and edge-level embedding augmentations into a joint similarity that also refines the adjacency, and use high-confidence K-means pseudo-labels to apply a weight modulation function that up-weights hard positives and down-weights hard negatives in the InfoNCE loss.
Key results
- On CORA, RAGC reaches 78.74% ACC and 60.62% NMI, versus 77.07% ACC for the second-best HSAN.
- On BAT, RAGC reaches 79.77% ACC, 2.65 points above HSAN.
- Under Gaussian attribute noise (sigma 0.1 to 0.3), RAGC shows an average degradation of 11.36%, versus 19.03% for DCRN and 20.61% for SCGC.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use RAGC instead of HSAN | Attributed graph clustering · CORA | ACC: 77.07±1.56 -> 78.74±0.72 | Helped |
| Use RAGC instead of HSAN | Attributed graph clustering · BAT | ACC: 77.15±0.72 -> 79.77±1.29 | Helped |
| Use RAGC instead of HSAN | Attributed graph clustering · CITE | ARI: 47.05±1.12 -> 46.51±0.63 | Hurt |
| Add dynamic confidence-factor selection instead of static high-confidence sample selection (ablation) | RAGC · Six benchmark datasets | clustering performance | Helped |
| Add HCA (hybrid node- and edge-level augmentation) (ablation) | RAGC · Six benchmark datasets | clustering performance | Helped |
| Add CSADA instead of classical InfoNCE loss (ablation) | RAGC · Six benchmark datasets | clustering performance | Helped |
| Add attribute Gaussian noise and masking in HCA | RAGC · Graph clustering under attribute noise | average degradation: -19.03 -> -11.36 | Helped |
| Add CSADA contrastive weighting | SCGC and HSAN · Six benchmark datasets | clustering performance | Helped |
| Vary β and γ across the tested ranges | RAGC · Six benchmark datasets | clustering performance | No clear effect |
| Increase noise standard deviation or feature-mask ratio excessively | RAGC · HCA parameter sensitivity | ACC | Hurt |
Limitations
- Authors note potential challenges on highly sparse and dynamic graphs.
- Baseline results are taken directly from original papers rather than re-run under identical conditions.
- Several hyperparameters (beta, gamma, sigma_N, r) are tuned per dataset in a table.
Experimental setup
Datasets: CORA CITESEER AMAP BAT EAT UAT
Benchmarks: CORA CITESEER AMAP BAT EAT UAT
Metrics: ACC NMI ARI F1
Baselines: DAEGC SDCN DFCN ARGA AGE NCLA CCGC GraphLearner GDCL ProGCL HSAN SCGC
Builds on: InfoNCE SCGC HSAN GraphLearner DCRN
Closest papers by meaning
- THESAURUS: Contrastive Graph Clustering by Swapping Fused Gromov-Wasserstein Couplings (AAAI 2025 · similarity 0.88)
- Multi-View Graph Clustering via Node-Guided Contrastive Encoding (ICML 2025 · similarity 0.87)
- Community-Invariant Graph Contrastive Learning (ICML 2024 · similarity 0.86)
- Deep Contrastive Graph Learning with Clustering-Oriented Guidance (AAAI 2024 · similarity 0.86)
- Graph Contrastive Learning with Joint Spectral Augmentation of Attribute and Topology (AAAI 2025 · similarity 0.86)
- Contrastive Auxiliary Learning with Structure Transformation for Heterogeneous Graphs (AAAI 2025 · 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.