Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

SOE: Fachverband Physik sozio-ökonomischer Systeme

SOE 24: Group Dynamics

SOE 24.4: Vortrag

Donnerstag, 14. März 2013, 16:30–16:45, H37

Prediction and predictability in systems with fat-tail distribution — •Jose M. Miotto and Eduardo G. Altmann — Max Planck Institute for the Physics of Complex Systems, Dresden, Germany

Availability of big databases of social media has triggered a wave of studies in such systems. Many different systems have shown to follow similar statistical behaviour (fat-tailed distributions) and to be modelled through similar models. Here we investigate the implications of these observations to prediction and predictability, into what extent it is possible to make a good prediction, and what are the factors that limit its quality. We focus our study in social systems in which many items are competing for a share of public attention; usually in these systems the distribution of activity among its items is fat-tailed, a feature that poses a big challenge to predictability. A paradigmatic case of study is the YouTube website: we collected a huge unbiased database of videos' activity time series, in which we tested some simple prediction schemes, and we report rigorous statistical measures of their performance and reliability. We analysed the possibilities of having a prediction with or without information about the previous popularity of a video, and we studied the role that fluctuations have in the performance measures.