Um método para acelerar a detecção de comunidades em redes complexas esparsas

• Autor(a) principal: Carnivali, Gustavo Simões
• Data de Publicação: 2019
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Biblioteca Digital de Teses e Dissertações do LNCC
• Texto Completo: https://tede.lncc.br/handle/tede/282
• topic: Ciência de redes; Teoria dos grafos; Método de Louvain; Clustering; CNPQ::CIENCIAS EXATAS E DA TERRA::MATEMATICA::ALGEBRA::LOGICA MATEMATICA

A graph is a structure composed of a set of objects (vertices) that may be connected by edges, which establish a relationship between a pair of vertices. A common problem that finds great applicability in several areas is the attempt to find communities in this structure. The community detection problem in graphs aims at finding sets of vertices that have some common relationship. In the literature dedicated to this subject, an optimization algorithm, called Louvain Method (ML), stands out as an effective and fast option for community detection in large scale complex networks. This work proposes an analysis about the graph structure targeting the community detection problem and, as a main result, an effective method for decreasing the processing time experienced by ML. In this context, the Fast Method of Grouping Vertices (MRAV in Portuguese) is proposed to quickly generate an initial solution (i.e., a first community set) that will be used by the ML. MRAV only replaces the first application of the constructive ML algorithm. With this modification, the processing time of ML can be reduced, at the cost of a slight reduction in the quality of the detected communities. The method was experimentally validated using real and artificial networks and it was observed that the proposal is more indicated for networks of low average degree, i.e., sparse networks. For these networks of low average degree, the proposal obtained a reduction in the process time of the ML of 47% with negligible reductions on the quality of the detected community structure.