Research map / Graph learning / Graph self-supervised learning
Attribute-Missing Multi-view 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 AMMGC, a multi-view graph clustering method for attribute-missing graphs that imputes missing node features via Dirichlet-energy-based neighborhood propagation, aligns cross-view structure, and uses high-confidence pseudo-label guidance.
Problem
Multi-view graph clustering when node attributes are entirely missing for some nodes in some views, where existing imputation methods are single-view, not clustering-specific, and ignore neighbor relations.
Approach
Iteratively imputes missing node attributes by Euler-scheme propagation over the normalized adjacency (minimizing Dirichlet energy), then combines a Laplacian-filtered GCN-style encoder with a cross-view structure alignment loss and a Student-t soft-assignment loss supervised by reliable K-means pseudo-labels.
Key results
- On CORA at missing rate 0.9, AMMGC reaches 68.94 ACC, 51.20 NMI, 45.87 ARI, outperforming the second-best method by 18.48% ACC.
- On AMAC, ACC drops only 2.66% from missing rate 0.3 to 0.9 (56.04 to 53.38).
- Context-aware imputation outperforms zero, random and mean imputation across four datasets (ablation at missing rate 0.9).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add context-aware imputation (CAI) (ablation) | AMMGC · CORA, CITESEER, AMAC, and WIKI (missing rate 0.6) | ACC: 55.74±0.72 -> 72.05±0.82; 63.54±1.12 -> 66.87±1.13; 51.42±1.67 -> 55.26±0.89; 41.87±0.64 -> 47.99±0.65 | Helped |
| Add dual structure consistency (DSC) (ablation) | AMMGC · CORA, CITESEER, AMAC, and WIKI (missing rate 0.6) | ACC: 70.55±1.79 -> 72.05±0.82; 55.72±3.69 -> 66.87±1.13; 53.83±0.72 -> 55.26±0.89; 43.91±1.75 -> 47.99±0.65 | Helped |
| Add high-confidence guidance (HCG) (ablation) | AMMGC · CORA, CITESEER, AMAC, and WIKI (missing rate 0.6) | ACC: 71.65±0.80 -> 72.05±0.82; 66.18±1.13 -> 66.87±1.13; 54.63±0.76 -> 55.26±0.89; 47.60±0.49 -> 47.99±0.65 | Helped |
| Use context-aware imputation instead of zero, random, or mean imputation (ablation) | AMMGC · CORA, CITESEER, AMAC, and WIKI (missing rate 0.9) | clustering performance | Helped |
| Increase attribute missing rate from 0.3 to 0.9 | AMMGC · CORA, CITESEER, AMAC, and WIKI | ACC: 73.64±1.37 -> 68.94±0.46; 69.08±0.75 -> 55.25±0.29; 56.04±0.46 -> 53.38±0.95; 55.32±1.18 -> 38.08±0.85 | Mixed |
Surprising findings
- Several baselines (e.g., CONVERT on WIKI at missing rate 0.6/0.9) collapse to near-trivial clustering, and HSAN runs out of memory on AMAC.
Limitations
- Evaluated only on four graph datasets with two views constructed from Gaussian-noise perturbation of one view.
- Paper does not report computational cost or scalability beyond noting an OOM failure for a baseline.
Experimental setup
Datasets: CORA CITESEER AMAC WIKI
Benchmarks: CORA CITESEER AMAC WIKI
Metrics: ACC NMI ARI
Baselines: SAT FPGM T2-GNN CONVERT CCGC SCGC HSAN
Builds on: SAT CCGC FeatProp (FPGM) GCN
Closest papers by meaning
- Attribute-Missing Graph Clustering Network (AAAI 2024 · similarity 0.92)
- Adaptive Feature Imputation with Latent Graph for Deep Incomplete Multi-View Clustering (AAAI 2024 · similarity 0.92)
- Scalable Attribute-Missing Graph Clustering via Neighborhood Differentiation (ICML 2025 · similarity 0.91)
- Incomplete and Unpaired Multi-View Graph Clustering with Cross-View Feature Fusion (AAAI 2025 · similarity 0.90)
- A Non-parametric Graph Clustering Framework for Multi-View Data (AAAI 2024 · similarity 0.90)
- Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding (AAAI 2024 · similarity 0.90)
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.