A Seed Growth Algorithm for Local Clustering in Complex Networks

Abstract

A seed growth algorithm based on a local connectivity rule for cluster formation in complex networks is introduced. That accompanies with the cluster normalization algorithm, the parameter determination process, and the pseudocluster inference process, forming the coherent algorithms and formalizing the categories within the realm of the cluster space. The prime clusters can be extracted from the cluster space, so that the overlapping complexity of clusters is confined to the prime clusters. To decide unequivocally whether the coherent algorithms are efficient, we have to simulate on the overlapping stochastic block networks. Our simulation shows that the dice coefficient of the prime cluster corresponding to the overlapping target cluster is 0.978± 0.024 on average. It decodes the underlying meaning that the coherent algorithms can efficiently search out the prime clusters containing almost the same nodes as the overlapping clusters. It provides a firm foundation for a simulation on the Caenorhabditis elegans neuronal network, unraveling the neurons DD04, PDB, VA08, VB06, VB07, VD07, and VD08 lying in the major overlap of clusters, among them the ablation of DD04 and PDB in biological experiments has shown to result in a pronounced loss of controllability of motor behavior.