Karmaşık ağlarda komün tarama

Complex networks have become very popular since the last decade. They allow modeling a given system by representing its components and their relationships with nodes and links, respectively. One of the most prominent sub domains in complex network analysis is community detection, which consists in searching cohesive subgroups in complex networks. The researchers use the networks which are generated artificially in order to test their community detection algorithms. In this study, we search the effect of the realism level of those computer generated artificial networks on algorithms community detection performance. We first analyze the properties of generated networks with the model proposed by Lancichinetti et al., which is supposed to be the most realistic until now. We propose a modification to further improve the level of realism and study their consequences in terms of topology. Our modification improves significantly the level of realism and especially the robustness of the model. We then apply a representative panel of eleven community detection algorithms on generated networks with the original model and its modified versions. The analysis of performance shows that the relative value of algorithms is generally not affected by the level of realism, however significant differences emerge when considering individual algorithms. In particular, an increased level of realism makes the task of identifying communities clearly more difficult, causing a significant drop in performance for all algorithms.