The Identification of Communities in the Cyberspace of Social Networks Based on Average Measure of Similarity

Abstract

Cyberspace is an important area of social activity of people, which is connected with the turnover of information in communication networks: social networks, regional, departmental, corporate. Cyberspace unites citizens of many countries and cultures that collect and use a variety of information. One of the most important problems of the cyberspace of social networks is the definition of communities in cyberspace. Communities in the cyberspace of networks are characterized by a large number of compounds between participants. Detection and analysis of the community allows us to study the stability and the cyberspace of social networks. The following clustering methods are used in graph models: modularity-based, centrality-based, clique percolation-based, hierarchical-based, spectral-based clustering. The standard algorithms for finding communities in the cyberspace of social networks are: single linkage clustering algorithm; spatial clustering based on density applications with noise; Jarvis-Patrick clustering algorithm; k-means clustering algorithm; mean-shift clustering algorithm; partitioning around medoids (PAM); minimum spanning tree-based clustering algorithm; spectral clustering algorithm.In community discovery methods, results depend on the similarity measure of participants. As a metric for the search for communities of graph models, measures of centrality by: degree, closeness, betweenness, radiality, eccentricity, PageRank, Katz, status, eigenvector. Similarity measures are applied based on correlation coefficient, on cosine coefficient, the Dice and Jaccard coefficients.We proposed, as measure for similarity apply the averaged coefficient on basis of correlation coefficient, cosine coefficient, the Dice and Jaccard coefficients. For graph models the algorithm based on averaged coefficient is proposed.A comparative analysis of the results of clustering is performed. The analysis showed that the choice of averaged coefficient as the metric gives the best result for clustering in comparison with standard clustering methods based on centrality and modularity.