Degree-corrected distribution-free model for community detection in weighted networks

Area of Science:

• Network Science
• Statistical Modeling
• Data Mining

Background:

• Real-world weighted networks exhibit complex structures with varying node degrees.
• Existing distribution-free models and degree-corrected stochastic block models have limitations in capturing these complexities.
• Accurate community detection in weighted networks is crucial for understanding social dynamics.

Purpose of the Study:

• To propose a novel degree-corrected distribution-free model for weighted social networks.
• To extend existing models to incorporate node degree variations and weighted edges.
• To develop a robust method for community detection in weighted networks with latent structures.

Main Methods:

• A degree-corrected distribution-free model is developed for weighted networks.
• Spectral clustering algorithm is employed for model fitting.
• Theoretical analysis of consistent estimation and performance under different weight distributions is provided.
• A generalized modularity measure is proposed for weighted networks.

Main Results:

• The proposed model effectively captures latent structural information in weighted networks.
• The spectral clustering algorithm provides consistent estimation.
• The generalized modularity measure accurately identifies network communities.
• Experimental results demonstrate superior performance compared to uncorrected methods.

Conclusions:

• The degree-corrected distribution-free model offers a significant advancement for analyzing weighted social networks.
• The proposed method enhances community detection accuracy and provides a more effective modularity measure.
• This work provides a robust framework for understanding complex weighted network structures.