Abstract:During the process of rainstorm, obtaining the real-time rainstorm disaster information, especially the effects and consequences of the disaster, is of great significance to the timely response and evaluation of the rainstorm disaster. Combining the latent Dirichlet allocation(LDA)and the supported vector machine(SVM), a mining model forrainstorm disaster information based on social media is proposed. The rainstorm event on April 11, 2019 is selected as an example, and 10 015 Weibo texts about “Shenzhen rainstorm” are collected. Four types of Weibo text themesare identified, and after its secondary mining, six kinds of disaster information as well as the temporal and spatial differences are obtained. The results show that the proposed mining model forrainstorm disaster information performs well on the identification of disaster information of the Weibo texts, and extract six kinds of disaster information of Shenzhen rainstorm: traffic congestion, casualty, building collapse, ponding, power failure and water outage. The mining and analysis results of rainstorm disaster information can accurately reflect the development status of the disaster. The temporal results of disaster information reflect its real-time development, and spatial results indicate that the area with concentrated Weibo texts agrees with that with occurrence of the disaster. The proposed mining approach for rainstorm disaster information may provide the basic and real-time information support for emergency rescue and relief work.