Tracking spatio-temporal variation of geo-tagged topics with social media in China: A case study of 2016 hefei rainstorm

Abstract Tracking how discussion geotagged topics evolve over time in social media has the potential to provide emergency information in natural disasters. In the traditional evolution of social media topics of disaster events, there are some problems, such as the topic of spatial subcategory aggregation is rarely discussed. Due to the behavior patterns of the focus groups in disaster events, it is inevitable to focus on the transfer and evolution of the focus topics of individuals in different periods before, during and after the disaster. In this study, we establish a framework to collect social media data from Sina Weibo, extract and analyze valuable information for emergency management. By using the case of 2016 Hefei rainstorm and flood disaster in China, this paper proposes a novel attempt to introduce latent Dirichlet allocation (LDA) model into density-based spatial clustering of applications with noise (DBSCAN) for exploring topics of spatial sub-categories, and quantitatively analyze the temporal evolution characteristics of fine-grained topics related to natural disasters by combining LDA topics and Markov transition probability matrix. The results indicate that before the rainstorm, people are advised to prepare for rainy travel, and they show attention for traffic inconvenience caused by pavement flooding. During the rainstorm, the continuous rainfall has had a serious impact on the traffic situation between Hefei and other cities. Floods broke out in certain areas, and the action of flood fighting and rescue is imminent. After the rainstorm, the weather gradually improved, and post-disaster reconstruction is particularly important. This study has the potential to promote effective policies and will allow decision-makers to respond quickly in the mitigation of damage caused by natural disasters.