Enhancing Community Detection in Complex Networks Using the Definition of Membership Degree for each nod

To recognize the societies in complex networks, Iso Fdp algorithm has used the combination of IsoMap algorithm so as to reduce the linear dimensions and clustering algorithm based on FDP density. One of the problems of this method is selecting the clusters’ centers or societies using the decision graph. The reason is that it would face problems in overlapped societies. Hence, the current study has investigated a method to overcome the problem. To this end, non-parametric clustering algorithm of Meanshift was used for clustering. Unlike the FDP algorithm, the suggested algorithm does not need to determine the cluster`s centers. Moreover, this algorithm has tried to solve the problem of cluster overlapping and membership problem of nods in each cluster through defining the vector of nods membership degree. Additionally, it has tried to choose the maximum difference to determine the clusters` centers and different weighing to their neighbors through defining different kernel functions. In this research for investigating the suggested method and the previous methods, 5 graphs of real networks of Football, Dolphins, Les Miserable and artificial ones like LFR and GN were used. In addition, for evaluating the results, NMI and Modularity criteria were used. The results of experiments using the above criteria on real and artificial networks indicated that the suggested method of recognizing the societies in comparison to the previous methods had remarkably improved.