|
Abstract:
|
Twitter is one of the most popular social networking websites, with millions of active users. These users tweet in real time about significant events in their lives, so it is possible to use the stream of tweets to detect major events. In this paper, we investigate tweets from Purdue University during two events - a local shooting on Jan 21 2014, and the Superbowl on Feb 2 2014. We use a Hidden Markov Model-based change point algorithm to detect the beginning of each event, and to break them up into sub-events. We then use a parametric bootstrap technique to find confidence intervals on the location of the change points. Finally we pick out the representative tweets between change points, thus giving an interpretation of the sub-events.
|
