Abstract:
This paper proposes a large language model-enhanced visual analytics approach for community detection to address the challenge that traditional community detection methods struggle to effectively identify and segment the numerous overlapping port nodes in maritime networks. Integrating visual analytics with ex-pert domain knowledge strengthens users’ interactive understanding of overlapping community structures. Additionally, the emergent reasoning and generation capabilities of large language models improve the ef-ficiency of the analysis and provide interpretability for the results. The maritime network, constructed from ship trajectory data, is segmented from node and link perspectives. Users explore overlapping community structures interactively through multiple coordinated views, including data overview, maritime network, port characteristics, link matrix, and community projection. Based on this approach, a visual analytics sys-tem enhanced by large language models is designed and implemented. The system supports the community detection process through guided exploration and result interpretation powered by the language model, of-fering decision support for analyzing global trade trends and enhancing the resilience of maritime trans-portation. Case studies on community structure detection and critical port identification demonstrate the effectiveness and practicality of the proposed method in uncovering complex maritime network community structures.