User Interaction Based Bursty Topic Model for Emergency Detection

When an emergency suddenly occurs, people usually share information and feelings in the social network. Therefore, it is of great significance to detect emergencies by analyzing and mining messages posted by users. Considering social network contains a mass of user interaction behavior, in this paper, we proposed a novel bursty topic model for emergency detection, named User Interaction based Bursty Topic Model (UIBTM). To overcome the problem of short text sparsity and ambiguity, UIBTM first uses comment texts and the amount of users liking the microblog to enrich the semantic of microblog, then generates the bursty topic model for bursty topic discovery and emergency detection. Comprehensive experiments on the dataset of Sina Microblog show that UIBTM can effectively overcome the sparsity of short text and detect emergencies efficiently.