An Efficient Online Event Detection Method for Microblogs via User Modeling

Detecting events in microblog is important but still challenging. As tweet stream is a mixture of user interests and external events, its expensive to distinguish them. Existing methods are ineffective since they ignore user interests or only model interests and events on a fixed dataset without scalability. In this paper, we introduce an online learning model User Modeling Based Interest and Event Topic Model (UMIETM). UMIETM (1) exploits user modeling's information to discover events, which usually capture attentions from users with different interests, and (2) treats the arriving data as stream and run the detection in online learning style. Furthermore, UMIETM can handle dynamic increased vocabulary in tweet stream. The UMIETM is verified on the real dataset which spans one year and contains 16 million tweets, and it outperforms state-of-the-art models in quantitative.CPCI-S(ISTP)huangwaleking@gmail.com; pekingchenwei@pku.edu.cn; citlmzhang@163.com; tjwang@pku.edu.cn329-341993