Studies on human mobility, including Bike Sharing System Analysis, have expanded over the past few years. They aim to give insight into the underlying urban phenomena linked to city dynamics and generally rely on data-mining tools to extract meaningful patterns from the huge volume of data recorded by such complex systems. This paper presents one such tool through the introduction of a family of generative models based on Poisson mixtures to automatically analyse and find temporal-based clusters in Origin/Destination flow-data. Such an approach may provide latent factors that reveal how regions of different usage interact over time. More generally, the proposed methodology can be used to cluster edges of temporal valued-graphs with respect to their temporal profiles and is thus particularly suited to mine patterns in dynamic Origin/Destination matrices commonly encountered in the field of transport. An in-depth analysis of the results of the proposed models was carried out on two months of trips data recorded on the Vélib׳ Bike-Sharing System of Paris to validate the approach.