Efficient and Effective One-Step Multiview Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Computer Science

Background:

• Multiview clustering algorithms show great promise but struggle with large datasets due to cubic complexity.
• Existing methods often use a two-stage approach, leading to suboptimal clustering results.

Purpose of the Study:

• To propose an efficient and effective one-step multiview clustering (E2OMVC) method.
• To address the scalability and suboptimality issues of current multiview clustering techniques.

Main Methods:

• Constructing smaller similarity graphs based on anchor graphs for each view.
• Generating low-dimensional latent features to form latent partition representations.
• Fusing latent representations and employing a label discretization mechanism for direct clustering indicator acquisition.

Main Results:

• The E2OMVC method directly obtains clustering indicators with reduced time complexity.
• A joint framework couples latent information fusion and the clustering task for improved results.
• Experimental results show comparable or superior performance against state-of-the-art methods.

Conclusions:

• The proposed E2OMVC method offers an efficient and effective solution for large-scale multiview clustering.
• The one-step approach and joint framework contribute to enhanced clustering performance.
• The method demonstrates scalability and effectiveness in practical applications.