Nonparametric resampling of random walks for spectral network clustering

Area of Science:

• Network Science
• Data Science
• Computational Social Science

Background:

• Complex network analysis often uses parametric resampling for statistical significance.
• Assessing graph clustering and community partition robustness is crucial.
• Existing methods may not fully capture network structural features.

Purpose of the Study:

• To propose a non-parametric resampling method for complex networks.
• To replicate structural features of networks using unbiased random walks.
• To enhance community detection algorithms.

Main Methods:

• Non-parametric resampling of the transition matrix from unbiased random walks.
• Application of bootstrapping techniques to synthetic and real-world modular networks.
• Evaluation of the ensemble of replicates for improving spectral clustering.

Main Results:

• The proposed method effectively replicates structural features of complex networks.
• The ensemble of replicates improves the performance of spectral community detection algorithms.
• The technique is validated on both synthetic and real-world modular networks.

Conclusions:

• Non-parametric resampling of transition matrices offers a robust approach for network analysis.
• This method enhances the reliability and accuracy of community detection in complex networks.
• The technique provides a valuable tool for understanding network structure and modularity.