提示路径对ChatGPT辅助应急预案编制效果的影响作用  

Influence of prompt paths on effect of ChatGPT aided emergency plan formulation

作　　者：孔胜利 杜崇瀚 钱继发 KONG Shengli;DU Chonghan;QIAN Jifa(School of Urban Construction and Safety Engineering,Shanghai Institute of Technology,Shanghai 201418,China)

机构地区：[1]上海应用技术大学城市建设与安全工程学院,上海201418

出　　处：《中国安全生产科学技术》2024年第8期27-32,共6页Journal of Safety Science and Technology

基　　金：上海应用技术大学中青年教师科技人才发展基金项目(ZQ2022-8)。

摘　　要：为探索人工智能辅助应急预案编制的可行性,有效提升预案编制效率,基于ChatGPT文本投喂,采用案例提示和规则提示2种方式对大语言模型辅助预案编制技术进行探索和尝试,并通过ROUGE方法对不同提示路径下应急预案的文本性能进行评估。研究结果表明:2种路径下均能形成相关预案的内容文本,在文本性能上,案例提示法整体优于规则提示法,特别的对于结构性较强的文本内容,单案例提示明显优于多案例提示;但对于内容灵活性较强的部分,多案例提示的文本性能能够明显提升。研究结果可为采用类似技术开展应急预案智能辅助编制时的提示路径选择提供数据支持和实践参考。In order to explore the feasibility of artificial intelligence aided emergency plan formulation and effectively improve the efficiency of plan formulation,based on ChatGPT text feeding,two methods of case prompt and rule prompt were used to explore and try the large language model aided plan formulation technology,and the ROUGE method was used to evaluate the text performance of emergency plans under different prompt paths.The results show that the content text of relevant plans can be formed under the two paths.In terms of text performance,the case prompt method is superior to the rule prompt method as a whole,especially for the text content with strong structure,and the single case prompt is obviously superior to the multi-case prompt.However,the text performance of partial multi-case prompt with strong content flexibility can be significantly improved.The research results can provide data support and practical reference for the selection of prompt path when using similar technologies to carry out the intelligence aided formulation of emergency plans.

关 键 词：ChatGPT 应急预案 文本精确度 ROUGE评估 

分 类 号：X913.2[环境科学与工程—安全科学]