Non-negative Tensor Factorization Based Bi-Clustering on Anchor Graphs

Area of Science:

• Machine Learning
• Data Science
• Artificial Intelligence

Background:

• Non-negative matrix factorization (NMF) is widely used for clustering.
• Current multi-view clustering methods using NMF often focus on decision-level fusion.
• Existing NMF-based methods struggle to leverage data-level relationships between views and lack interpretability.

Purpose of the Study:

• To propose a novel multi-view clustering model addressing limitations of existing NMF-based approaches.
• To enhance clustering by integrating both data-level and decision-level fusion.
• To improve the interpretability of multi-view clustering models.

Main Methods:

• Developed a new multi-view clustering model based on non-negative tensor factorization (NTF).
• Employed multi-level fusion (data-level and decision-level) for view-consistent labels.
• Utilized NTF to decompose a tensorized anchor graph into probabilistic anchor and data indicator tensors for enhanced interpretability.

Main Results:

• The proposed NTF-based model effectively achieves view-consistent labels in multi-view clustering.
• The method demonstrates superior performance compared to existing approaches in extensive experiments.
• The probabilistic interpretation enhances the model's transparency and understanding.

Conclusions:

• The novel NTF model offers a more effective and interpretable approach to multi-view clustering.
• Multi-level fusion is crucial for exploiting inter-view relationships at both data and decision levels.
• The probabilistic decomposition provides a foundation for more transparent and reliable clustering outcomes.