A Degree-based Distributed Label Propagation Algorithm for Community Detection in Networks

Community detection is an important way to understand the functions and characteristics of complex systems. The Label Propagation Algorithm (LPA) is a community detection algorithm of which complexity is close to linear. Due to the randomness and instability of the algorithm, it is difficult to get good results with this algorithm. This paper proposes a degree-based label propagation algorithm and parallelizes the algorithm based on the GraphX component in Spark platform (D-disLPA). Using WeChat empirical network data, this paper adopts modularity as an evaluation index to compare D-disLPA with traditional label propagation algorithms. The experimental results show that the D-disLPA can effectively solve the non-convergence problem in traditional algorithms and improve the stability and accuracy of community detection. At the same time, this distributed algorithm is able to satisfy the requirements of large-scale community detection.