International audience This paper, based on recorded geolocated Twitter messages, attempts to identify urban patterns by separating various modes of urban events from urban routine. We set out to identify relationships between types of user, types of event and the transmission of messages on the Twitter social network for Paris and the Île-de-France region. To do so, we used a graph method making it possible to combine spatial and temporal simplifications. This method highlights the spatiotemporal variations of nodes and flows over long time series of spatial situations. Beyond preferential trajectories of Twitter users, Parisian clusters appear: hotspots, crossing points, clusters during a specific event. Dynamics graph can reveal the " ephemeral city " and the behaviours development of Twitter users in Paris.