5D-1
Using location entropy and observed tweeting behavior to identify large events
○Asif Hossain Khan Md（The University of Tokyo），岩井将行，瀬崎 薫（東大）
In this paper we present a noble method of identifying the occurrence of major events involving city dwellers. Instead of using data collected in controlled laboratory environment as done by most contemporary research, we have used Tweeter’s open platform to monitor the tweeting behavior of over 450 users for four months. Analyzing this vast data we have extracted a set of features to identify each user’s normal threshold of tweets which varies dynamically depending on the time of the day. We have also used location entropy, which measures the average expected tweets done from that location at a particular time of the day and a particular day of the week. Using these thresholds, we have devised a model that can predict occurrence of major events experienced by the Tweeter users in real time. To the best of our knowledge, this is the first effort in identifying the change in human behavior in response to the occurrence of large events.