Spectral Clustering Community Detection Algorithm Based on Point-Wise Mutual Information Graph Kernel.

Area of Science:

• Network science
• Data mining
• Computational complexity

Background:

• Traditional spectral clustering methods struggle to capture complete network structural information.
• Accurate community detection is crucial for understanding complex network dynamics.

Purpose of the Study:

• To develop an advanced spectral clustering algorithm (PMIK-SC) for comprehensive network structural analysis.
• To introduce an efficient algorithm (BI-CNE) for determining the optimal number of communities in networks.

Main Methods:

• Constructing a point-wise mutual information (PMI) graph kernel to reconstruct network proximity.
• Utilizing eigendecomposition and eigenvector clustering on the Laplacian matrix for network partitioning.
• Implementing Bayesian inference with Monte Carlo sampling in BI-CNE for community number estimation.

Main Results:

• The proposed PMIK-SC algorithm demonstrates superior accuracy and stability in community detection compared to existing methods.
• The BI-CNE algorithm significantly outperforms other methods in detection speed and accuracy for estimating community numbers.
• PMIK-SC effectively captures complete structural information, overcoming limitations of traditional spectral clustering.

Conclusions:

• PMIK-SC offers a robust and accurate approach to network community detection by leveraging PMI graph kernels.
• BI-CNE provides an efficient and reliable solution for determining the number of communities, enhancing spectral clustering applications.
• The combined approach advances the field of network analysis and community detection algorithms.