A novel model of bursts in event sequences

When we focus on events analysis, such as studying the event burstiness and monitoring the event trends, we have to face a great number of similar events that had happened in the history, i.e. event sequences, usually spanning more than 10 years. Burst detection is a popular technique of sequence analysis. Today there are several burst models and detection algorithms based on different burst definitions, producing very different results. However, almost all of these definitions are difficult to clearly define the beginning and end, as well as intensity of the burst. In this paper, we reconsider the burst definition and propose an efficient detection approach for event sequences, which can be further utilized to other temporal sequences or data streams. As a sample application, we present the burst model for event sequences since the end of WWII (1946~2010) collected from the event lists in Wikipedia. Finally, we show the results comparison of our model and three popular ones in terms of rationality, and two case studies.