There are many subtle community structures in social networks that naturally form based on user interests and behavior patterns. Identifying these communities is important for public opinion guidance and precision marketing. However, traditional particle swarm optimization algorithms are prone to getting stuck in local optima during community discovery, and their fitness evaluation methods are relatively simple. To address these issues, this study proposes an improved Particle Swarm Optimization (MPSO) algorithm. This algorithm mainly introduces three aspects of improvement: firstly, designing a weighted fitness function that comprehensively considers modularity and community size balance, with the aim of enhancing community cohesion and improving the degree of separation between communities; Secondly, implement a dynamic adjustment mechanism for inertia weights, focusing on global exploration in the early stage and fine local search in the later stage; Thirdly, introduce an elite retention strategy to avoid the loss of good solutions. This study conducted experimental validation on multiple public datasets, including Zachary’s Karate Club, Douban Movie, as well as Twitter15 and Twitter16. The experimental results show that MPSO improves the average modularity by 12.3% compared to traditional particle swarm optimization algorithms; Its normalized mutual information index is also 8.7% higher than the Louvain algorithm; Meanwhile, the running time of this algorithm is reduced by about a quarter compared to genetic algorithms. The results of this study confirm that the improved particle swarm algorithm can effectively improve the performance of social network community discovery, providing a more efficient intelligent optimization method for mining hidden community structures.
Hui Du (Thu,) studied this question.