Localization of Unidentified Events with Raw Microblogging Data

Event localization is the task of finding the location of an event. Commonly, event localization using microblogging services, like Twitter, use con- tents of the messages and the geographical information associated with the messages. In this paper, we propose a novel approach called SPARE (SPAtial REconstruction) that bypasses the need for geographical or semantic information to localize tweets. We assume there are reference coordinates at known locations that scrape the microblog (tweet) counts in time and space (circular regions around the reference coordinate). The counts of tweets are aggregated which are then disaggregated to identify event patterns. The change in counts of tweets would be indicative of an event pattern. We show, using real data, that the change in counts of tweets is manifested as peaks. The peaks from multiple reference coordinates can be used as an input to trilateration techniques to pinpoint the location of an event. We introduce metrics to identify the quality of disaggregation of fine-grained data and examine techniques like filtering to improve accuracy of event location. The experimental results show that our method can identify the location of an event with high accuracy.