Research map / Graph learning / Graph self-supervised learning
SURER: Structure-Adaptive Unified Graph Neural Network for Multi-View 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 SURER, a structure-adaptive unified heterogeneous graph neural network that refines view-specific attribute graphs with graph autoencoders and links same-instance nodes across views to learn consensus representations for deep multi-view graph clustering.
Problem
Deep multi-view graph clustering relies on fixed, noisy graphs built from raw multi-view data and fuses views late, failing to propagate structure relations across views.
Approach
Jointly learn refined view-specific graphs via graph autoencoders (GCN encoder, feature decoder, inner-product graph decoder), connect same-instance representations across views into a unified heterogeneous graph, and propagate messages with a heterogeneous GNN before a KL-based self-supervised clustering layer.
Key results
- On 100leaves, SURER reaches ACC 0.9100 and NMI 0.9538, versus 0.8569 ACC for the best baseline CDIMC.
- On Hdigit, SURER reaches ACC 0.9834 versus 0.9478 for MFLVC on the strongest baseline comparison.
- On Scene15, SURER achieves ACC 0.4696 and NMI 0.4531, outperforming the best baseline SiMVC (ACC 0.4383).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use SURER instead of the strongest reported baseline | Multi-view clustering · 100leaves | ACC: 0.8569 -> 0.9100 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · BBCSports | ACC: 0.9430 -> 0.9632 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · Hdigit | ACC: 0.9478 -> 0.9834 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · VOC | ACC: 0.6113 -> 0.9786 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · MSRCv1 | ACC: 0.9238 -> 1.0000 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · Scene15 | ACC: 0.4383 -> 0.4696 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · Mfeat | ACC: 0.9540 -> 0.9765 | Helped |
| Use SURER instead of the strongest reported baseline | Multi-view clustering · Noisyminist | ACC: 0.4916 -> 0.5282 | Helped |
| Add the graph structure learning module (ablation) | SURER · Mfeat | clustering performance | Helped |
| Add unified heterogeneous graph construction linking views (ablation) | SURER · Mfeat | clustering performance | Helped |
| Combine graph structure learning with unified heterogeneous graph construction (ablation) | SURER · Mfeat | clustering performance | Helped |
Surprising findings
- Traditional shallow baselines such as MCGC achieve very high recall on some datasets while low precision, and several deep baselines underperform the simpler CDIMC on many datasets.
Limitations
- Hyperparameters lambda1 and lambda2 selected by grid search; sensitivity only analyzed on Mfeat.
- Noisyminist evaluated on a random 13k-instance subset rather than the full dataset.
- Ablation and sensitivity analyses reported only on Mfeat.
Experimental setup
Datasets: MSRCv1 BBCSports 100leaves Mfeat Scene15 VOC Hdigit Noisyminist
Benchmarks: MSRCv1 BBCSports 100leaves Mfeat Scene15 VOC Hdigit Noisyminist
Metrics: ACC NMI PUR ARI PRE REC F-score
Baselines: MCGC LMSC LMVSC CDIMC SiMVC CoMVC MFLVC DFP-GNN
Builds on: graph convolutional networks DEC-style KL clustering graph autoencoder
Closest papers by meaning
- Medusa: A Multi-Scale High-order Contrastive Dual-Diffusion Approach for Multi-View Clustering (CVPR 2025 · similarity 0.90)
- Global Graph Propagation with Hierarchical Information Transfer for Incomplete Contrastive Multi-view Clustering (AAAI 2025 · similarity 0.89)
- EASEMVC:Efficient Dual Selection Mechanism for Deep Multi-View Clustering (CVPR 2025 · similarity 0.89)
- A Non-parametric Graph Clustering Framework for Multi-View Data (AAAI 2024 · similarity 0.89)
- Attribute-Missing Multi-view Graph Clustering (CVPR 2025 · similarity 0.89)
- Homophily-Related: Adaptive Hybrid Graph Filter for Multi-View Graph Clustering (AAAI 2024 · similarity 0.88)
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.