Non-negative matrix factorization for overlapping community detection in directed weighted networks with sparse...

Area of Science:

• Complex Network Analysis
• Data Mining
• Graph Theory

Background:

• Analyzing complex networks is crucial for understanding system structure and function.
• Existing community detection methods often neglect valuable attribute information, focusing solely on network topology.
• Overlapping community detection is vital for networks where nodes can belong to multiple groups.

Purpose of the Study:

• To propose a novel attribute-information non-negative matrix factorization approach for detecting overlapping communities in directed weighted networks.
• To integrate sparse constraints and optimize an objective function that leverages both network topology and attribute data.
• To provide a rigorous convergence proof for the proposed algorithm's update rule.

Main Methods:

• Developed an attribute-information non-negative matrix factorization algorithm.
• Incorporated adaptive updates of the non-negative matrix using both topology and attribute information.
• Integrated graph regularization with sparsity constraints to preserve network geometry.
• Provided a strict convergence proof for the multiplication update rule.

Main Results:

• The proposed algorithm demonstrates superior effectiveness in detecting overlapping communities compared to existing methods on artificial and real-world networks.
• The study reveals the iterative process of system evolution towards convergence.
• Analysis highlights the impact of various variables on network detection accuracy.

Conclusions:

• The novel approach enhances overlapping community detection by effectively integrating network topology and attribute information.
• The findings offer valuable insights for constructing more robust and operable complex systems.
• The algorithm's proven convergence ensures reliability in complex network analysis.