基于大语言模型的群体性突发事件事理图谱构建与演化分析  

Construction and Analysis of the Event Evolution Graph for Mass Emergency Based on the Large Language Model

作　　者：黄少年 王焕然 李佩霖 Huang Shaonian;Wang Huanran;Li Peilin(Key Laboratory of Statistical Learning and Intelligent Computing in Hunan Province,Changsha 410205;School of Computer Science,Hunan University of Technology,Changsha 410205)

机构地区：[1]统计学习与智能计算湖南省重点实验室,长沙410205 [2]湖南工商大学计算机学院,长沙410205

出　　处：《情报杂志》2024年第12期161-169,共9页Journal of Intelligence

基　　金：国家社会科学基金一般项目“大数据背景下群体性突发公共事件应急决策模型与应用研究”(编号:21BTJ026)研究成果。

摘　　要：[研究目的]基于大语言模型自动构建群体性突发事件事理图谱,进行时序演化路径和关系演化路径的可视化分析,揭示群体性突发事件的发展脉络和演化规律,为应急决策提供理论和实践参考。[研究方法]构建群体性突发事件本体模型,设计基于ChatGPT的多轮提示模板实现零样本的实体、实体关系、事件、事件关系抽取,并设计评价指标对大模型信息抽取效果进行客观评估。采用Neo4j图数据库实现事理图谱的自动构建与可视化,并基于事理图谱查询进行群体性突发事件演化分析。[研究结论]基于大语言模型进行零样本的群体性突发事件信息抽取,既能发挥大语言模型上下文学习的智能性,又能节省时间和计算资源。其在各个任务上的F1值均优于传统的深度学习模型,验证了其在事理图谱自动构建中的可行性及在突发事件演化分析中的有效性,为大语言模型在突发事件应急决策中的应用提供了新方向。[Research purpose]Based on large language model,an Event Evolution Graph for Mass Emergency is constructed.Through the visual analysis of the time evolution path and the relational evolution path,the development law of mass emergency is revealed.The construction and analysis of event evolution diagrams provide theoretical and practical references for government emergency decision-making.[Research method]The ontology model of mass emergency is constructed,and the multi-round prompt template based on ChatGPT is designed to realize zero-shot information extraction.The evaluation criteria are designed to evaluate the effect of large model information extraction objectively.The Neo4j database is used to construct and visualize the event evolution graph.And then the evolution analysis of mass emergency is carried out by querying event evolution graph.[Research conclusion]The scheme for information extraction of mass emergency based on large language model not only gives play to the intelligence of contextual language learning of large language model,but also saves time and computational resources.The F1 value of each task is better than that of traditional deep learning model,which verifies its feasibility in the automatic construction of event graph and its effectiveness in the evolution analysis of emergency.It provides a new direction for the application of large language model in emergency decision making.

关 键 词：突发事件 大语言模型 本体模型 信息抽取 事理图谱 演化分析 

分 类 号：G250[文化科学—图书馆学]