基于社交文本的洪涝信息抽取与时空演变分析  

Information extraction and spatiotemporal evolution analysis of flood disaster based on social media text

作　　者：侯华伟 慎利[1] 贾嘉楠 徐柱[1] HOU Huawei;SHEN Li;JIA Jianan;XU Zhu(Faculty of Geosciences and Engineering,Southwest Jiaotong University,Chengdu 611756,China)

机构地区：[1]西南交通大学地球科学与工程学院,四川成都611756

出　　处：《地理与地理信息科学》2025年第2期1-9,共9页Geography and Geo-Information Science

基　　金：国家重点研发计划项目(2022YFB3904202);国家自然科学基金重大项目(42394063)。

摘　　要：以2018年寿光市水灾为例,基于深度学习和规则匹配相结合方法从微博数据中抽取关键灾情信息,通过时间序列分析提取洪涝事件的关键时间节点,利用核密度估计探索灾情的空间分布特征,应用HDBSCAN算法分析核心受灾区域;通过LDA算法对待响应点的文本进行主题分析,提取不同受灾区域面临的问题与需求;基于抽取的水情相关信息和DEM数据,利用HAND模型绘制洪水淹没范围,识别核心受灾区。实验结果表明,该框架的灾情信息抽取总体准确率达83%,高于其他对比方法,可为应急响应提供定向援助与资源配置的决策依据。Taking the flood disaster in Shouguang City in 2018 as an example,this paper collects relevant microblog data,and adopts a hybrid approach combining deep learning and regular matching to accurately and automatically extract key disaster information from social media text.On this basis,this paper extracts the key time nodes of flood events through time series analysis,uses kernel density estimation to explore the spatial distribution characteristics of flood disaster,and then uses HDBSCAN algorithm to analyze the core disaster-stricken area.The LDA algorithm is used to analyze the theme of the response information text to extract the problems and needs of different disaster-stricken areas.Finally,the extracted flood information and DEM data are used to map the flood inundation range using the HAND model.The results show that the overall accuracy of disaster information extraction is 83%,which is better than some traditional disaster information extraction methods.The framework can provide decision-making basis for targeted assistance and resource allocation for emergency response.

关 键 词：社交文本 深度学习 规则匹配 LDA模型 HAND模型 

分 类 号：P426.616[天文地球—大气科学及气象学] P208