Uma análise sobre a acurácia e a escalabilidade de algoritmos paralelos de detecção de comunidades em grafos

Abstract

Community detection is a type of topological analysis widely used in graph analysis in several fields such as social network analysis, bioinformatics and recommendation systems. The problem involves detecting components that have high internal density and low external density. Due to the rapid growth in the volume of data from a variety of applications and the wide use of this type of analysis, several researches in parallel and distributed approaches to solve the problem of community detection have emerged. Some algorithms are more popular, resulting in an extensive amount of research on optimizations for parallel processing. Other algorithms, which posses better accuracy results in tests, do not present the same level of research depth in their parallel and distributed versions. This research addresses the accuracy and scalability of three community detection algorithms. User guidelines are proposed based on the experiments results. In addition, the behavior of the parallel approaches is explored and possible improvements are proposed.