| 摘要: |
| 人工智能（AI）正成为推动人类迈入智能时代的决定性力量。近年来，ChatGPT、DeepSeek 等大语言模型（LLMs）因其在语言理解与推理任务中的高度通用性和智能化表现，引发广泛关注， 并推动了一场社会系统级的变革。在此背景下，如何将 LLMs 有效引入结构复杂、专业性强的政务 审查等领域，已成为人工智能应用研究中的重要挑战之一。海域使用论证审查是保障海域使用申请 和市场化出让海域使用权依法依规开展的关键审批环节。当前深圳市的审查模式存在效率低、重复 性高、人员投入大等问题，且由于跨部门协同工作缺乏统一标准，导致审查结果容易受人为因素影 响，难以保障其规范性和一致性。文章聚焦于“如何提升 LLMs 在多源结构化文本理解、合规性判定与审查意见生成中的推理能力与知识融合效果”这一核心科学问题，面向海洋管理场景构建了“海 域使用论证报告审查 AI 助手”模型框架。该框架融合大模型的问答推理能力与行业知识图谱，实现对政策条款、图表信息及第三方材料的多模态解析与语义一致性判断。研究还系统分析了大模型 在审查任务中的上下文保持能力、合规性边界识别及跨文档推理表现，提出了一套可迁移的模型优 化策略和评估体系。实验结果表明，所提出的框架在海域使用审查任务中相较于传统的人工审查具 有显著的效率提升，并在准确性方面取得了 90％以上的测试表现，验证了其在实际应用场景中的可行性与有效性。研究为海域使用审查工作的智能化、标准化与精准化提供了理论与技术基础，同 时为大语言模型在高语义控制任务中的政务审查与合规监测应用提供了具有可迁移性的参考框架。 |
| 关键词: 人工智能 海域使用论证 大语言模型 智慧海洋 合规性审查 |
| DOI：10.20016/j.cnki.hykfygl.2025.11.012 |
| 投稿时间：2025-05-06修订日期：2025-11-05 |
| 基金项目: |
|
| Intelligent Review Methods of Maritime Space Utilization Argumentation Reports Based on Large Language Models |
| ZHANG Huihui |
| Shenzhen Marine Development & Promotion Center |
| Abstract: |
| Artificial intelligence（AI）is becoming a decisive force in propelling humanity into the intelligent era．In recent years，large language models（LLMs）such as ChatGPT and DeepSeek have garnered widespread attention due to their high versatility and intelligent performance in language comprehension and reasoning tasks，driving a systemic societal transformation．In this context，how to effectively introduce LLMs into fields such as governmental review，which involve complex structures and specialized expertise，has become one of the key challenges in AI application research．The review of maritime space utilization argumentation reports is a critical approval process
to ensure that maritime space utilization applications and the market-based transfer of sea use rights are conducted in accordance with laws and regulations．The current review model in Shenzhen faces challenges，including low efficiency，high redundancy，and intensive personnel input．Additionally，the lack of unified standards for cross
department collaboration leads to subjective differences in review results，making it difficult to ensure consistency and standardization．This study focuses on the core scientific issue of“how to enhance the reasoning ability and knowledge integration of large language models in multi-source structured text understanding，compliance determination，and review opinion generation．”An“AI Assistant for Maritime Space Utilization Argumentation Reports Review”is developed for marine management scenarios．This framework integrates the question-answering and reasoning capabilities of large models with industry knowledge graphs，enabling multimodal analysis and semantic consistency judgment of policy terms，charts，and third-party materials．The study also systematically analyzes the performance of large models in review tasks，including contextual retention，compliance boundary identification，and cross-
document reasoning，and proposes a set of transferable model optimization strategies and evaluation systems．The results demonstrate that the proposed framework significantly enhances efficiency compared with traditional manual review in the task of maritime spatial use argumentation，achieving over 90％ accuracy in testing．This provides strong
evidence of its effectiveness in application scenarios．The study provides theoretical and technical foundations for
the intelligent，standardized，and precise enhancement of maritime space utilization reviews，while also offering a transferable reference framework for the application of large language models in governmental review and compliance monitoring tasks with high semantic control requirements． |
| Key words: Artificial intelligence,Maritime space utilization argumentation,Large language model,Smart
ocean,Compliance review |
