Multi-view clustering based on feature selection and semi-non-negative anchor graph factorization.

Area of Science:

• Machine Learning
• Data Mining
• Artificial Intelligence

Background:

• Multi-view clustering leverages diverse data perspectives for improved analysis.
• Existing anchor graph methods face limitations in large-scale data handling and label derivation.
• Current anchor selection and graph construction are often disjointed, impacting performance.

Purpose of the Study:

• To propose a unified framework for multi-view clustering integrating feature selection and anchor graph factorization.
• To enhance clustering accuracy and stability by addressing limitations of existing anchor graph techniques.
• To develop a method that directly yields clustering labels without post-processing steps like K-means.

Main Methods:

• Introduced Multi-view Clustering based on Feature Selection and Semi-Non-Negative Anchor Graph Factorization (MCFSAF).
• Unifies feature selection, anchor learning, and anchor graph factorization within a single framework.
• Employs tensor Schatten p-norm minimization for cross-view information discovery and semi-non-negative factorization for indicator matrix generation.

Main Results:

• MCFSAF demonstrated superior performance in comprehensive experimental evaluations.
• The unified framework effectively enhances clustering quality through synergistic anchor selection and graph learning.
• Direct label acquisition from the fused indicator matrix significantly improved clustering stability.

Conclusions:

• The proposed MCFSAF method offers a robust and efficient approach to multi-view clustering.
• Integrating feature selection and anchor graph learning in an embedding space improves clustering outcomes.
• Eliminating the need for additional K-means clustering enhances the stability and reliability of the results.