Unveiling the influencing factors of maritime accidents through data-driven approaches: leveraging large language model tools

IF 6.2 1区工程技术 Q1 ENGINEERING, INDUSTRIAL

Safety Science Pub Date : 2026-05-01 Epub Date: 2026-01-19 DOI:10.1016/j.ssci.2026.107116

Meng Wei , Ying Cui , Jiaguo Liu

{"title":"Unveiling the influencing factors of maritime accidents through data-driven approaches: leveraging large language model tools","authors":"Meng Wei , Ying Cui , Jiaguo Liu","doi":"10.1016/j.ssci.2026.107116","DOIUrl":null,"url":null,"abstract":"<div><div>This study investigates the influencing factors of maritime accidents using a data-driven approach integrated with large language models (LLMs). First, DeepSeek V3 is utilized to extract data on humans, vessels, management, environment, time, and space from maritime accident reports. Second, the N-K model is applied to analyze coupled risks, revealing that the coupled risk value of the four factors is the highest, with humans, vessels, and management being key factors, while the environmental factor exerts a significant impact in specific accidents. Further spatiotemporal analysis indicates phased temporal risks and spatially aggregated coastal risks. A significant nonlinear superposition effect is identified: high spatial risk amplifies the marginal contribution of temporal risk increases, while low-level coupling significantly reduces overall risk. The study culminates in proposing a dynamic collaborative governance framework. This framework leverages an integrated cognitive architecture, powered by LLMs, to transform unstructured data into actionable risk intelligence, enabling graduated intervention protocols and optimized resource allocation for enhanced maritime safety management.</div></div>","PeriodicalId":21375,"journal":{"name":"Safety Science","volume":"197 ","pages":"Article 107116"},"PeriodicalIF":6.2000,"publicationDate":"2026-05-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Safety Science","FirstCategoryId":"5","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S092575352600007X","RegionNum":1,"RegionCategory":"工程技术","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"2026/1/19 0:00:00","PubModel":"Epub","JCR":"Q1","JCRName":"ENGINEERING, INDUSTRIAL","Score":null,"Total":0}

引用次数: 0

Abstract

This study investigates the influencing factors of maritime accidents using a data-driven approach integrated with large language models (LLMs). First, DeepSeek V3 is utilized to extract data on humans, vessels, management, environment, time, and space from maritime accident reports. Second, the N-K model is applied to analyze coupled risks, revealing that the coupled risk value of the four factors is the highest, with humans, vessels, and management being key factors, while the environmental factor exerts a significant impact in specific accidents. Further spatiotemporal analysis indicates phased temporal risks and spatially aggregated coastal risks. A significant nonlinear superposition effect is identified: high spatial risk amplifies the marginal contribution of temporal risk increases, while low-level coupling significantly reduces overall risk. The study culminates in proposing a dynamic collaborative governance framework. This framework leverages an integrated cognitive architecture, powered by LLMs, to transform unstructured data into actionable risk intelligence, enabling graduated intervention protocols and optimized resource allocation for enhanced maritime safety management.

查看原文本刊更多论文

通过数据驱动的方法揭示海上事故的影响因素：利用大型语言模型工具

本研究采用数据驱动的方法与大型语言模型（llm）相结合，探讨了海上事故的影响因素。首先，利用DeepSeek V3从海上事故报告中提取人员、船舶、管理、环境、时间和空间等数据。其次，运用N-K模型对耦合风险进行分析，发现4个因素的耦合风险值最高，其中人、船舶和管理是关键因素，而环境因素对具体事故的影响较大。进一步的时空分析表明，分阶段的时间风险和空间聚集的沿海风险。高空间风险放大了时间风险边际贡献的增加，而低耦合显著降低了总体风险。该研究最终提出了一个动态协作治理框架。该框架利用由法学硕士提供支持的集成认知架构，将非结构化数据转换为可操作的风险情报，实现分级干预协议和优化资源分配，以增强海上安全管理。

本文章由计算机程序翻译，如有差异，请以英文原文为准。

求助全文

约1分钟内获得全文求助全文

来源期刊

Safety Science 管理科学-工程：工业

CiteScore

13.00

自引率

9.80%

发文量

335

审稿时长

53 days

期刊介绍： Safety Science is multidisciplinary. Its contributors and its audience range from social scientists to engineers. The journal covers the physics and engineering of safety; its social, policy and organizational aspects; the assessment, management and communication of risks; the effectiveness of control and management techniques for safety; standardization, legislation, inspection, insurance, costing aspects, human behavior and safety and the like. Papers addressing the interfaces between technology, people and organizations are especially welcome.