Title: Modeling Collective Anticipation and Response on Wikipedia

Abstract: The dynamics of popularity in online media are driven by a combination of endogenous spreading mechanisms and response to exogenous shocks including news and events. However, little is known about the dependence of temporal patterns of popularity on event-related information, e.g. which types of events trigger long-lasting activity. Here we propose a simple model that describes the dynamics around peaks of popularity by incorporating key features, i.e., the anticipatory growth and the decay of collective attention together with circadian rhythms. The proposed model allows us to develop a new method for predicting the future page view activity and for clustering time series. To validate our methodology, we collect a corpus of page view data from Wikipedia associated to a range of planned events, that are events which we know in advance will have a fixed date in the future, such as elections and sport events. Our methodology is superior to existing models in both prediction and clustering tasks. Furthermore, restricting to Wikipedia pages associated to association football, we observe that the specific realization of the event, in our case which team wins a match or the type of the match, has a significant effect on the response dynamics after the event. Our work demonstrates the importance of appropriately modeling all phases of collective attention, as well as the connection between temporal patterns of attention and characteristic underlying information of the events they represent.