View-Driven Multi-View Clustering via Contrastive Double-Learning.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Multi-view clustering aims to leverage information from multiple data sources.
• Existing deep learning methods struggle to balance consistency and diversity between views.
• A unified approach is needed to effectively integrate both aspects for improved clustering.

Purpose of the Study:

• To propose a novel deep learning method for multi-view clustering that effectively balances consistency and diversity.
• To enhance feature learning and clustering accuracy by integrating view-driven information and dual contrastive learning.
• To address the limitations of current methods that overemphasize either consistency or diversity.

Main Methods:

• Developed a view-driven multi-view clustering (VMC-CD) method.
• Employed a view-driven strategy to incorporate information from other views, promoting diversity.
• Implemented dual contrastive learning to align features and clustering results across views.

Main Results:

• The VMC-CD method demonstrated superior performance compared to state-of-the-art methods.
• Experimental results on three datasets validated the effectiveness of the proposed approach.
• The method successfully balanced consistency and diversity for better clustering outcomes.

Conclusions:

• The proposed VMC-CD method offers an effective solution for multi-view clustering.
• Dual contrastive learning and view-driven approaches significantly improve clustering quality.
• This work advances deep multi-view clustering by addressing the critical balance between information consistency and diversity.