Logo Logo
Switch Language to German
Schmid, Klaus Arthur; Frey, Christian; Peng, Fengchao; Weiler, Michael; Zufle, Andreas; Chen, Lei; Renz, Matthias (2016): TrendTracker: Modelling the motion of trends in space and time. IEEE 16th International Conference on Data Mining Workshops (ICDMW), 12-15 Dec. 2016, Barcelona, Catalonia, Spain.
Full text not available from 'Open Access LMU'.


Both the current trends in technology such as smart phones, general mobile devices, stationary sensors and satellites as well as a new user mentality of utilizing this technology to voluntarily share information produce a huge flood of geo-textual data. Such data includes microblogging platforms such as Twitter, social networks such as Facebook, and data from news stations. Such geo-textual data allows to immediately detect and react to new and emerging trends. A trend is a set of keywords associated with a time interval where the frequency of these keywords is increased significantly. In this paper, we investigate the dissemination of trends over space and time. For this purpose, we employ a four-step framework. In the first step, we employ existing solutions to mine a large number of trends. Second, for each trend we create a spatio-temporal dissemination model, which describes the motion of this trend over space and time. To model this dissemination, we employ a (flow-source, flow-destination, time, trend) tensor. In the third step, we cluster these trend-tensors, to identify groups of archetype trends. For each archetype, we aggregate all tensors of the same archetype, and employ a tensor factorization approach to describe this archetype by its latent features. As the fourth step, we propose an algorithm which can classify the trend-archetype of a new trend, in order to predict the future dissemination of this trend. In our experiments, we are able to show that the space of trends does exhibit clusters, each corresponding to a trend-archetype such as political trends, disaster trends and celebrity trends. We show that by identifying the trend-archetype of a trend, we can effectively predict the future of this trend.