An effective and efficient label initialization method based on similarity for community detection in networks (2201.06089v1)

Abstract: Identifying clusters or community structures in networks has become an integral part of social network analysis. Though many methods were proposed, the label propagation algorithm (LPA) is a popular computationally efficient method with running time linear. However, the LPA provides different combination of communities on the same network due to the randomness in LPA. Many improvements have been proposed to tackle this stability problem by eliminating the randomness. This paper put forward an improvement to the standard LPA by proposing a label initialization method based on link similarity. The similarity is measured based on the connection strength between two nodes. The method is tested on real and synthetic measures to analyze the performance.