Combined method to detect communities in graphs of interacting objects

Abstract

АbstractIn this paper, we propose and implement a method for detecting intersecting and nested communities in graphs of interacting objects of diff erent natures. For this, two classical algorithms are taken: a hierarchical agglomerate and one based on the search for k-cliques. The combined algorithm presented is based on their consistent application. In addition, parametric options are developed that are responsible for actions with communities whose sizes are smaller than the given k, and also with single vertices. Varying these parameters allows us to take into account diff erences in the topology of the original graph and thus to correct the algorithm.The testing was carried out on real data, including on a group of graphs of a social network, and the qualitative content of the resulting partition was investigated. To assess the diff erences between the integrated method and the classical algorithms of community detections, a common measure of similarity was used. As a result, it is clearly shown that the resulting partitions are signifi cantly diff erent. We found that for the approach proposed in the article the index of the numerical characteristic of the partitioning accuracy, modularity, can be lower than the corresponding value for other approaches. At the same time, the result of an integrated method is often more informative due to intersections and nested community structure.A visualization of the partition obtained for one of the examples by an integrated method at the fi rst and last steps is presented. Along with the successfully found set of parameters of the integrated method for small communities and cut off vertices in the case of social networks, some shortcomings of the proposed model are noted. Proposals are made to develop this approach by using a set of parametric algorithms.

Klasterisasi Pahlawan Nasional Indonesia Berdasarkan Daerah Asal Menggunakan Algoritma Community Detection

“…Secara definisi, modularitas adalah karakteristik numerik dari sebuah jaringan, tingkat akurasi dari dibuatnya suatu partisi [14]. Algoritma ini diajukan oleh Newman juga, sebagai metode lain klasterisasi hierarki aglomeratif [15].…”