RBF-SC: A Fast Community Detection Technique Using Radial Basis Functions

We propose an optimized algorithm for detecting communities in complex networks. We apply the radial basis functions to the corresponding similarity matrices according to the different networks. The structure of the network is exploited to develop an efficient matrix algorithm for the resulting systems. The proposed algorithm is utilized with the Spectral Clustering method. We adopt a two-stage training paradigm and propose a simple approach to finding the communities of the complex networks: (1) finding the proper Radial Basis Function to construct the weighted matrix; (2) finding the optimal shaping parameter with respect to the conditioning of the corresponding weighted matrices. Our approach achieves state-of-the-art accuracy on different types of networks. The numerical experiments demonstrate the benefits of incorporating the shaping parameter into radial basis functions. The results of simulations conducted on multiple data sets indicate that the optimized algorithm achieves exceptional performance, particularly for large data sets. A software package to generate the RBF networks and perform SC-RBF algorithms can be downloaded from GitHub.