VLP: A Label Propagation Algorithm for Community Detection in Complex Networks

Sharon Boddu, Maleq Khan, Mais Nijim

Abstract: Community detection is a commonly encountered problem in social network analysis and many other areas. A community in a graph or network is a subgraph containing vertices that are closely connected to other vertices within the same subgraph but have fewer connections to the other vertices. Community detection is useful in analyzing complex systems and recognizing underlying patterns and structures that govern them. There are several algorithms that currently exist for community detection, ranging from simple and fast approaches, such as the label propagation algorithm (LPA), to more complex and time-consuming methods, such as the state-of-the-art Louvain method. We propose a new method called vector label propagation (VLP), which is a generalization of the LPA approach. The VLP algorithm significantly enhances the quality of the detected communities compared to LPA while being much faster than the Louvain method. For example, on the Twitter network, VLP has a normalized mutual information (NMI) score of 0.82, while LPA has an NMI score of 0.47. With rigorous experimentations, we demonstrate that the VLP algorithm is significantly faster than state-of-the-art algorithms such as Louvain and Infomap. On the Twitter network, VLP is 2.8 times faster than Louvain.

External IDs:dblp:conf/asunam/BodduKN24