TwiTracker: Detecting and Extracting Events from Twitter for Entity Tracking

Among the existing social platforms, Twitter plays a more and more important role in social sensing due to its real-time nature. In particular, it reports various types of events occurred in the real world, which provides us the possibility of tracking entities of interest (e.g., celebrities, organizations) in real time via event analysis. Hence, this paper presents TwiTracker, a system for obtaining the timelines of entities on Twitter. The system uses Twitter API and keyword search to collect tweets containing the entities of interest, and combines event detection and extraction together to extract elements including activities, time, location and participants. Online incremental clustering is further applied to fuse extraction results from different tweets to remove redundant information and enhance accuracy. Echarts is used to visualize the dynamic trajectory of each entity under tracking. For evaluation, we take Golden State Warriors, a famous NBA team, as well as the stars in the team as the experimental objects to compute their timelines, and compare the experimental results with the ground truth data hunted from the Internet, which demonstrates TwiTracker is effective for tracking entities and can provide information that is not covered by newswires.