Research map / Learning theory and algorithms / Clustering
Discriminatively Fuzzy Multi-View K-means Clustering with Local Structure Preserving
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 DFMKLS, a fuzzy multi-view K-means that adds a between-cluster separation term and a local kNN structure-preservation term, optimized with multiplicative updates.
Problem
Multi-view K-means only enforces compactness within clusters and ignores separability between clusters, and it relies on global structure while neglecting local neighborhood structure.
Approach
Combines within-cluster compactness, maximized between-cluster center distances, a relaxed fuzzy (nonnegative, non-one-hot) assignment matrix, and a kNN graph Laplacian term into one objective for multi-view clustering, solved by alternating multiplicative updates.
Key results
- On 3Sources, DFMKLS reaches Accuracy 0.8107 and NMI 0.7096 versus 0.7396 and 0.7066 for the best prior baselines (FMK/DFMK) in accuracy and NMI respectively.
- On BBC, DFMKLS reaches Accuracy 0.8672 and NMI 0.7341, outperforming CGL (0.8068 accuracy).
- On WebKB, DFMKLS reaches Accuracy 0.9610 and NMI 0.7276, above RMKMC (0.9515 accuracy).
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add local structure preserving to DFMK (ablation) | DFMK · 3Sources | Accuracy: 0.7396 -> 0.8107 | Helped |
| add local structure preserving to DFMK (ablation) | DFMK · BBC | Accuracy: 0.6496 -> 0.8672 | Helped |
| add local structure preserving to DFMK (ablation) | DFMK · WebKB | Accuracy: 0.9467 -> 0.9610 | Helped |
| add local structure preserving to DFMK (ablation) | DFMK · NUS WIDE | Accuracy: 0.2331 -> 0.2453 | Helped |
| add discriminative cluster separation to FMK (ablation) | FMK · 3Sources | Accuracy: 0.7396 -> 0.7396 | No clear effect |
| add discriminative cluster separation to FMK (ablation) | FMK · BBC | Accuracy: 0.5547 -> 0.6496 | Helped |
| add discriminative cluster separation to FMK (ablation) | FMK · WebKB | Accuracy: 0.8687 -> 0.9467 | Helped |
| add discriminative cluster separation to FMK (ablation) | FMK · NUS WIDE | Accuracy: 0.2252 -> 0.2331 | Helped |
| increase α above 0.01 | DFMKLS · 3Sources, BBC, WebKB, NUS WIDE | NMI | Hurt |
| vary K across tested values | DFMKLS · WebKB, NUS WIDE | clustering validity indexes | No clear effect |
Surprising findings
- Too large local-structure weight alpha sharply degrades clustering indices, especially NMI.
- Several baselines such as Spec-Pair and MVASM achieve very low NMI on BBC (about 0.01-0.02), despite reasonable accuracy.
Limitations
- Objective is non-convex and only converges to a local minimum.
- Per-iteration complexity is quadratic in the number of samples n.
Experimental setup
Datasets: 3Sources BBC WebKB NUS WIDE
Benchmarks: 3Sources BBC WebKB NUS WIDE
Metrics: Accuracy F-score NMI Precision
Baselines: BSV Concat MultiNMF Spec-Pair Spec-Cent RMKMC MVASM EMKMC GMC CGL FMK DFMK
Builds on: K-means fuzzy multi-view k-means RMKMC
Closest papers by meaning
- Pseudo Multi-view K-means Clustering (AAAI 2026 · similarity 0.91)
- Sharper Error Bounds in Late Fusion Multi-view Clustering with Eigenvalue Proportion Optimization (AAAI 2025 · similarity 0.87)
- Towards Resource-friendly, Extensible and Stable Incomplete Multi-view Clustering (ICML 2024 · similarity 0.85)
- Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding (AAAI 2024 · similarity 0.85)
- Partial Multi-View Clustering via Self-Supervised Network (AAAI 2024 · similarity 0.85)
- Fast Incomplete Multi-view Clustering by Flexible Anchor Learning (ICML 2025 · similarity 0.84)
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.