融合ERNIE的自然灾害舆情事理图谱构建及次生衍生事件探测研究  

Research on the Construction of Natural Disaster Public Opinion Event Evolutionary Graph Integrated with ERNIE and Detection of Secondary Derivative Events

作　　者：李诗轩 王璐[1] 沈愿 陈烨 周禾深 Li Shixuan;Wang Lu;Shen Yuan;Chen Ye;Zhou Heshen(School of Safety Science and Emergency Management,Wuhan University of Technology,Wuhan 430070;School of Information Management,Nanjing University,Nanjing 210023;School of Information Management,Wuhan University,Wuhan 430072)

机构地区：[1]武汉理工大学安全科学与应急管理学院,武汉430070 [2]南京大学信息管理学院,南京210023 [3]武汉大学信息管理学院,武汉430072

出　　处：《情报杂志》2025年第3期128-138,共11页Journal of Intelligence

基　　金：国家自然科学基金青年项目“情景视角下基于事理图谱的领域突发事件溯因与演化推理方法研究”(编号:72204194);大学生创新创业训练计划项目“突发事件背景下基于事件演化推理技术的应急决策方法研究”(编号:S202410497254)研究成果。

摘　　要：[研究目的]基于规则模版和大语言模型构建面向自然灾害的网络舆情事理图谱,结合情感分析进行舆情演化分析,进而实现次生衍生事件探测,为突发事件舆情管理与预警提供理论和实践参考。[研究方法]分别利用规则模板和预训练模型ERNIE提取关系事件对,并利用K-means和BERTopic进行事件泛化,构建自然灾害网络舆情抽象事理图谱。同时利用情感词典进行情感演化分析,结合同类灾害事件的相似度计算,构建同类事件的网络舆情预测事理图谱。[研究结果/结论]自然灾害舆情抽象事理图谱能够分析事件的因果与顺承关系,揭示舆情演化过程中的关键事件。预测事理图谱可以辅助次生衍生事件探测,实现舆情预警。对比发现基于预训练模型ERNIE结合BERTopic构建的图谱抽象事件较少,事件集中于中心节点,而规则模板结合K-means得到的图谱抽象事件较多,关系较为分散。[Research purpose]This study aims to construct a network public opinion event evolutionary graph for natural disasters based on rule templates and big language models.By integrating sentiment analysis,this study seeks to analyze public opinion evolution and detect secondary and derivative events,providing theoretical and practical references for public opinion management and early warning of sudden events.[Research method]This study utilizes rule templates and the pre-trained model ERNIE to extract relational event pairs.K-means and BERTopic are employed for event generalization to construct an abstract event evolutionary graph for natural disaster-related public opinion.Sentiment dictionaries are used for sentiment evolution analysis,and similarity calculations of similar disaster events are utilized to build predictive event evolutionary graphs for similar events.[Research result/conclusion]The abstract event evolutionary graph for natural disaster-related public opinion can analyze causal and consequential relationships within events,revealing key events in the evolution of public opinion.The predictive event evolutionary graph can assist in detecting secondary and derivative events,enabling public opinion early warning.A comparison reveals that the abstract event graph constructed using the pre-trained model ERNIE combined with BERTopic results in fewer abstract events,with events concentrated around a central node,while the graph constructed using rule templates combined with K-means results in more abstract events with more dispersed relationships.

关 键 词：网络舆情 自然灾害 事理图谱 预训练模型ERNIE 情感演化分析 次生衍生事件探测 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]