Consensus multi-view spectral clustering network with unified similarity.

Area of Science:

• Computer Science
• Machine Learning
• Data Mining

Background:

• Multi-view spectral clustering requires learning a consensus representation from heterogeneous data.
• Existing methods often construct affinity matrices separately, limiting unified similarity learning.
• Lack of explicit consistency enforcement in embedding representations leads to suboptimal clustering.

Purpose of the Study:

• To propose a deep multi-view spectral clustering network for effective consensus representation learning.
• To address limitations in unified similarity and embedding consistency in existing methods.
• To improve clustering performance through enhanced representation learning.

Main Methods:

• Developed a deep spectral embedding learning framework integrating data for unified similarity across views.
• Constructed a spectral mapping network to extract common embedding representations.
• Employed local structure-constrained contrastive learning to align spectral embedding representations and enforce consistency.

Main Results:

• The proposed framework successfully learns unified similarity across multiple views.
• Local structure-constrained contrastive learning effectively aligns spectral embeddings.
• Comparative experiments on eight public datasets demonstrate the algorithm's superiority and effectiveness.

Conclusions:

• The proposed deep multi-view spectral clustering network effectively learns consensus representations.
• Unifying similarity and aligning embeddings significantly improves clustering performance.
• The method offers a superior approach for multi-view spectral clustering tasks.