ChatGPT支持下的震例分析报告生成方法研究  

ChatGPT-based Generative Methods of Earthquake Case Analysis Reports

作　　者：杨瑞琦 李忠[1,3,4] 黄建平 霍禹萌 朱可欣 YANG Ruiqi;LI Zhong;HUANG Jianping;HUO Yumeng;ZHU Kexin(Institute of Intelligent Emergency Information Processing,Institute of Disaster Prevention,Sanhe 065201,China;School of Information Engineering,Institute of Disaster Prevention,Sanhe 065201,China;School of Emergency Management,Institute of Disaster Prevention,Sanhe 065201,China;Hebei Key Laboratory of Resource and Environment Disaster Mechanism and Rish Monitoring,Sanhe 065201,China;National Institute of Natural Disaster Prevention,Ministry of Emergency Management,Beijing 100085,China)

机构地区：[1]防灾科技学院智能应急信息处理研究所,河北三河065201 [2]防灾科技学院信息工程学院,河北三河065201 [3]防灾科技学院应急管理学院,河北三河065201 [4]河北省资源环境灾变机理及风险监控重点实验室,河北三河065201 [5]应急管理部国家自然灾害防治研究院,北京100085

出　　处：《防灾科技学院学报》2024年第4期95-103,共9页Journal of Institute of Disaster Prevention

基　　金：河北省研究生专业学位精品教学案例(库)建设项目(KCJPZ2023060);国家重点研发计划政府间国际科技创新合作项目(2023YFE17300)。

摘　　要：生成式大语言模型目前已成为人工智能在自然语言处理领域的研究热点,它可以帮助人类从繁重的工作和重复劳动中解放出来。以ChatGPT为例,探讨生成式大语言模型在震例分析中的应用,降低地震科研工作者的劳动强度,提高工作效率。实验发现,在设定的聊天情景下,ChatGPT可以理解输入的自然语言内容,生成与之相符的答复结果;通过不断地与ChatGPT进行地震领域的知识交流,借助ChatGPT出色的前后学习机制,向ChatGPT输入以往震例分析报告的语言风格和报告模板,使得ChatGPT能够学习地震知识,生成与当前震例报告风格一致的震例分析报告;通过对比不同报告模板生成的分析报告内容,人工剔除报告中的错误信息,选择优质的震例分析报告,从而提高了报告的实用价值。这拓宽了ChatGPT的应用领域,为地震工作者提供了更加便利的服务。Generative large language models,which can help human beings free themselves from heavy work and repetitive labor,have become a research hotspot in the field of the artificial intelligence applied in natural language processing.Taking ChatGPT as an example,this paper discusses the application of generative large language model in earthquake case analysis with an aim to reduce the labor intensity and improve the work efficiency of earthquake researchers.The experimental results show that ChatGPT can understand the input natural language content and generate the corresponding reply results under the set chat situation.Through continuous knowledge exchange with ChatGPT in the field of seismology and on the basis of ChatGPT's excellent pre-and post-learning mechanism,we input the language style and report templates of previous earthquake case analysis reports into ChatGPT.This enables ChatGPT to learn seismic knowledge and generate earthquake case analysis reports that are consistent with the input report style.By comparing the content of analysis reports generated by different report templates and manually eliminating the error information in the reports,high-quality earthquake case analysis reports are selected,thus improving their practical value.This widens the application field of ChatCPT and provides a more convenient service for earthquakeworkers.

关 键 词：生成式大语言模型 ChatGPT 震例分析报告 语言风格 

分 类 号：P315.71[天文地球—地震学] P315-39[天文地球—固体地球物理学]