Community detection has wide-ranging applications in semantic networks, biological networks, and social network analysis. Swarm intelligence algorithms have been extensively applied to this problem, resulting in notable progress. One such approach is EP-WOCD (Evolutionary Population Whale Optimization for Community Detection) (Feng et al. in Appl Intell 50:2503–2522, 2020), a robust algorithm based on enhanced whale population behaviors and an evolutionary population strategy. Building upon EP-WOCD, we propose a new algorithm, IEP-WOCD (Improved Evolutionary Population Whale Optimization for Community Detection) which addresses the risk of premature convergence by introducing strategies to maintain population diversity across generations. These strategies help avoid local optima and enhance the potential for global optimization. Furthermore, IEP-WOCD extends the search space exploration during the final stage, thereby reducing the risk of losing the best-found solution, as observed in EP-WOCD. Experiments were conducted to compare IEP-WOCD with several state-of-the-art algorithms, including including EP-WOCD on artificial and real-world social networks. The experimental results demonstrate that IEP-WOCD achieves superior performance in both effectiveness and stability.