Détection et analyse de communautés dans les réseaux sociaux

Abstract

To model some complex systems, it is appropriate to use mathematical structures called graphs or networks. The problem posed by graphs is to detect communities. The goal is to understand the structure by detecting a partition. Many algorithms have been used to solve this problem. One of the methods we used is the percolation of cliques. It is based on finding a group of nodes more closely connected to each other than other nodes in the network. But this algorithm goes through several steps which makes it very slow in execution of a large number of nodes. The second method is the Louvain algorithm, which is currently one of the best algorithms in terms of complexity for calculating communities on very large graphs. This method has the particularity of implementing a local "gluttonous" optimization method of modularity. Finally we have created a heuristic. It simulates large real graphs, which we have obtained from random numbers of nodes and edges. It allows the detection and graphic visualization of communities. We also tested it on real graphs, obtained from social networks. A comparative study was carried out to observe the quality of partitioning as well as the detection time of the communities by the proposed algorithms