What a Difference a Day Makes: The Potential of Different Event Detection Methods with Twitter Data for Varying Times Spans

Increasingly ICT are used to coordinate, document, and communicate about collective action, such as protest events, strikes, relevant official proceedings or media coverage thereof. Thus ICT enable - or at least facilitate - collective action, they also leave data trails that allow researchers to reconstruct collective action in its temporal structure and its dynamics. It has been shown that data collected from internet platforms provide researchers with new insights into human behavior online as well as offline. As ICT become more prominent in efforts to coordinate, organize and communicate about collective action they might also offer us new insights into the temporal structure and the dynamics of collective action - at least in those cases in which collective offline found a strong echo online. One such channel is the microblogging service Twitter. As has been previously shown changes in the frequency and dynamics of Twitter messages on a given topic allow researchers to identify key events that happened during a respective time span. Twitter messages can thus be used to reconstruct and analyze the temporal structure of the phenomena users chose to tweet about. In this paper we advance this argument by testing various event detection approaches to illustrate their varying potential for the detection of events during short time spans (of up to two days) and longer time spans (of up to four months). We show that there is no one size fits all solutions but that researchers have to consciously choose a method that fits their topic and the time span that they are interested in. We analyze all Twitter messages by the 80.000 most active Twitter users in Germany commenting on the highly controversial project Stuttgart 21. We identified 142.200 Twitter messages posted between 1 June 2010 and 19 October 2010 containing the phrase #s21.