Dual-Force Community Evolution: Integrating Peer Influence and Interest Similarity in Social Networks

Understanding how communities evolve requires accounting for both social influence and individual interests. In this paper, the Community Evolution Framework, which leverages the Louvain algorithm for initial community detection and extends it with a dynamic evolution mechanism was proposed. The framework integrates the peer influence and interest similarity through a tunable scoring function, allowing community memberships to be updated iteratively until convergence. To enhance interpretability, this paper furthers incorporate the large language models to generate thematic explanations of evolved communities based on interest distributions. The experiments conducted on a large-scale student social network dataset demonstrate that the proposed framework achieves strong modularity while capturing realistic membership reorganization patterns. The results highlight that balancing peer pressure and personal interests produces more cohesive yet adaptable communities. This work provides a quantitative and interpretable approach to studying the community evolution, with potential applications to broader the network science and the social computing tasks.