GrSrNMF: dynamic community detection with graph and symmetry bi-regularized non-negative matrix factorization

Area of Science:

• Network Science
• Data Mining
• Machine Learning

Background:

• Dynamic community detection is crucial for analyzing evolving networks in fields like social media and e-commerce.
• Existing evolutionary clustering methods often focus on change points rather than direct modeling of network evolution.
• Graph regularization in dynamic networks can suffer from over-smoothing and sensitivity to network noise.

Purpose of the Study:

• To propose a novel dynamic community detection framework, Graph and Symmetry Bi-regularized Non-negative Matrix Factorization (GrSrNMF).
• To address the over-smoothing issue and improve the modeling of network evolution patterns.
• To effectively identify community structures and variations in community numbers over time.

Main Methods:

• Developed the GrSrNMF framework incorporating graph and symmetry bi-regularization.
• Applied the model to analyze both synthetic and real-world dynamic network data.
• Compared GrSrNMF performance against state-of-the-art evolutionary clustering and graph regularization methods.

Main Results:

• GrSrNMF successfully identifies community structures and adapts to changing community numbers across network snapshots.
• The method mitigates the over-smoothing problem, preserving distinct node characteristics.
• GrSrNMF demonstrates superior performance compared to existing methods on various network datasets.

Conclusions:

• GrSrNMF provides an effective framework for dynamic community detection, capturing network evolution patterns.
• The proposed method enhances the analysis of dynamic networks by improving upon traditional graph regularization techniques.
• GrSrNMF offers a robust approach for understanding temporal changes in complex network structures.