Detecting Live Events by Mining Textual and Spatial-Temporal Features from Microblogs

摘要: As microblogging services on the mobile devices are widely used, microblogs can be viewed as a kind of event sensor to perceive dynamic behaviors in city. In particular, detecting live events microblogs, such mass gathering, emergencies, etc., help understand what happened from point view people who present. For identifying large number short and noisy paper builds generative probabilistic model named ST-LDA cluster whose semantics, time space similar into same topic, then determines topics by an HMM-based method. The conducts experiments real weibo.com. Experimental results show that our method detect more accurately completely than LDA-based TimeLDA-based