Variational Graph Generator for Multiview Graph Clustering.

Area of Science:

• Machine Learning
• Data Mining
• Graph Theory

Background:

• Multiview graph clustering (MGC) methods are crucial for analyzing complex multiview data with inherent graph structures.
• Existing MGC techniques struggle to simultaneously leverage both consensus graph information and view-specific feature details.

Purpose of the Study:

• To develop an advanced MGC method that effectively utilizes both common and specific information across multiple views.
• To address the limitations of current methods in integrating consensus graph information and view-specific features.

Main Methods:

• A novel variational graph generator is proposed to extract common information by inferring a variational consensus graph.
• A graph encoder and multiview clustering objective are employed to learn graph embeddings, integrating view-common and view-specific graphs with features.
• The Information Bottleneck (IB) principle is used to analyze the uncertainty of the inferred consensus graph.

Main Results:

• The proposed Variational Graph Generator for Multiview Graph Clustering (VGMGC) demonstrates superior performance.
• Extensive experiments confirm VGMGC outperforms current state-of-the-art methods in multiview graph clustering tasks.
• Theoretical analysis validates the rationality of VGMGC by examining consensus graph uncertainty.

Conclusions:

• VGMGC offers a robust framework for multiview graph clustering by effectively fusing diverse information sources.
• The method provides a significant advancement in leveraging both shared and unique characteristics within multiview graph data.
• The public availability of the source code facilitates further research and application of VGMGC.