Resumen
Social media is increasingly being used to obtain timely flood information to assist flood disaster management and situational awareness. However, since data in social media are massive, redundant, and unstructured, it is tricky to intuitively and clearly obtain effective information. To automatically obtain clear flood information and deduce flood development processes from social media, the authors of this paper propose an event-based and multi-level modeling approach including a data model and two methods. Through the hierarchical division of events (division into spatial object, phase, and attribute status), the flood information structure (including time, space, topic, emotion, and disaster condition) is defined. We built an entity construction method and a development process deduction method to achieve the automatic transition from cluttered data to orderly flood development processes. Taking the flooding event of the Yangtze and Huai Rivers in 2020 as an example, we successfully obtained true flood information and development process from social media data, which verified the effectiveness of the model and methods. Meanwhile, spatiotemporal pattern mining was carried out by using entities from different levels. The results showed that the flood was from west to east and the damage level was positively correlated with the number of flood-related social media texts, especially emotional texts. In summary, through the model and methods in this paper, clear flood information and dynamic development processes can be quickly and automatically obtained, and the spatiotemporal patterns of flood entities can be examined. It is beneficial to extract timely flood information and public sentiments towards flood events in order to perform better disaster relief and post-disaster management.