Self-Grouping Multi-Network Clustering.

Area of Science:

• Network science
• Data mining
• Machine learning

Background:

• Joint clustering of multiple networks improves accuracy over individual network clustering.
• Existing methods often assume a single, shared clustering structure across all networks.
• This assumption is limiting for real-world applications with diverse network data distributions.

Purpose of the Study:

• To develop a method that can simultaneously group and cluster multiple networks.
• To address the limitation of existing methods by accounting for diverse data distributions and underlying network groups.
• To improve multi-network clustering performance by identifying and leveraging shared structures within network groups.

Main Methods:

• Proposed ComClus, a novel method for simultaneous network grouping and clustering.
• ComClus utilizes node clusters as network features to differentiate between network groups.
• Network grouping and clustering are integrated and mutually reinforced within the learning process.

Main Results:

• ComClus effectively groups networks based on their underlying structures.
• Simultaneous grouping and clustering lead to improved performance compared to traditional methods.
• Experimental evaluations on synthetic and real datasets validate the method's effectiveness.

Conclusions:

• The proposed ComClus method offers an effective solution for joint multi-network clustering with diverse data distributions.
• Automatic network grouping enhances the accuracy and applicability of multi-network clustering.
• ComClus provides a robust framework for analyzing complex, multi-network data.