基于BERTopic的河南暴雨灾害微博舆情时空演化与主题结构研究

Spatiotemporal Evolution and Topic Structure of Weibo Public Opinion on the Henan Rainstorm Disaster Based on BERTopic

摘要: 为刻画重大暴雨灾害事件中公众信息需求与情绪表达的时空演化规律，文章以河南暴雨灾害相关微博为研究对象，构建“文本主题–时间演化–空间分布”一体化分析框架。基于网络爬虫获取微博文本数据13,509条，并对其中可定位至河南省内的5654条样本进行地理标准化处理；在此基础上，采用BERTopic模型完成语义嵌入、降维聚类与主题表示学习，提取灾害舆情主题并刻画其跨时间阶段的强度变化与空间分异特征。结果表明：1) 灾害舆情呈现显著的核心–边缘主题结构，以“应急救援与灾情进展”为核心主题，并与“捐赠互助、公共服务、次生灾害与预警、责任讨论与反思”等外围主题形成同心层级关系；2) 主题热度随灾害过程呈阶段性波动，整体表现为“突发爆发–持续发酵–回落沉淀”的生命周期特征，不同主题在爆发期与恢复期的主导性存在差异；3) 空间上，舆情强度在省内呈现以郑州为核心的集聚与外扩格局，并表现出随距离增加而减弱的扩散特征。研究可为暴雨灾害情景下的风险沟通策略、信息发布节奏优化与跨区域协同治理提供数据支撑与决策参考。

Abstract: To characterize the spatiotemporal dynamics of public information needs and sentiment expressions during a severe rainstorm disaster, this study takes Weibo posts related to the Henan rainstorm disaster as its research subject and constructs an integrated analytical framework of “textual themes-temporal evolution-spatial distribution”. A total of 13,509 Weibo posts were collected via web crawling, among which 5654 geotagged posts within Henan Province were standardized for spatial analysis. On this basis, the BERTopic was employed to perform semantic embedding, dimensionality reduction, clustering, and topic representation learning, enabling the extraction of disaster-related topics and the quantification of topic intensity across time stages and locations. The results show that: 1) disaster public opinion exhibits a clear core-periphery structure, with “emergency rescue and situation updates” as the core topic, surrounded by peripheral topics such as “donations and mutual aid, public services, secondary hazards and warnings, and responsibility discussions”; 2) topic intensity varies by disaster phases, following a lifecycle pattern of “outbreak, sustained fermentation, post-event decline”, with shifts in dominant topics between the outbreak and recovery stages; and 3) spatially, public opinion intensity within the province presents an agglomeration and outward expansion pattern centered on Zhengzhou, exhibiting a distance decay diffusion feature that weakens with increasing distance. These findings provide data support and decision-making references for risk communication strategies, optimize information release rhythms, and cross-regional collaborative emergency governance under rainstorm disaster scenarios.