Large-scale community detection in social networks

Abstract

There are various community detection algorithms which that have been developed. Among them, Louvain method is the most widely used algorithm because of its simplicity and good performance. The goal of this project is to improve an existing parallel implementation of community detection algorithm based on Louvain method that works on multiple GPU. This project empirically studies existing partitioning methods, memory and running time optimization. As the result of the studies, a new partitioning method was proposed to decrease the running time of overall algorithm. The functionality was also expanded by allowing weighted network as input. In addition, the running time of modularity computation was also improved.