Large language model enhanced maritime ship registration: The implications in advancing intelligent maritime governance

As a foundational component of maritime governance, international ship registration (ISR) involves rule-intensive language understanding and multi-step procedural reasoning. The advancement of generative artificial intelligence (AI) presents a novel opportunity to transform rule-intensive regulatory workflows by embedding large language models (LLMs) into registration processes. In this paper, an LLM-enhanced ISR framework is proposed to integrate domain knowledge with supervised fine-tuning methods. This framework can support structured procedural reasoning, required document verification, and compliance-oriented interactions in ISR tasks. First, a data engineering method is proposed to construct domain datasets from regulatory texts and administrative practices. This method enables procedural rules and compliance constraints to be encoded explicitly. Second, the performance of LLMs with different sizes is systematically evaluated across representative ISR task categories using full-parameter and parameter-efficient fine-tuning strategies. Finally, a real-world ISR case study from the Hainan Free Trade Port in China is used to validate the practical applicability of the proposed framework. Results indicate that domain-adapted LLMs can effectively enhance rule interpretation consistency and reduce repetitive explanation tasks. Failure mode analysis further reveals a clear trade-off between model efficiency and governance reliability. These findings provide methodological guidance and policy implications for deploying generative AI in maritime governance.