A precise knowledge of internal dynamics of a complex socio-technical system is generally difficult to infer, as such a system involving human agents will necessarily present unpredictable emergent properties on the one hand, and precise behavioral data is most of the times impossible to obtain on the other hand (for example the realization of developed user surveys, generally used to elaborate and calibrate transportation systems models with an aim of improvement of the level of service, is heavy to implement and has a certain cost). We propose an hybrid modeling approach to unveil internal mechanisms of a particular type of system that are bike-sharing transportation systems, combining poor quality survey data and raw data on system evolution, fed into an hybrid statistical and agent-based model. We work on the real system of Paris (Vlib) with 1 year of recent data on system dynamics. The method should be easily generalized to other kind of systems, as the core principle of hybrid modeling has already been well documented in the literature (see e.g. recent works in archeology [Crabtree and Kohler, 2012]). The model is internally (statistical robustness) and externally (calibration through cross-validation) validated and parameter space is explored to have a precise knowledge of the behavior of the model. We can draw several conclusions from the exploration of the model, as the influence of some user parameters on the level of service (proposing user-based scenarii for an improvement of the system), but also more advanced conclusions as the indirect retrieving of discrete choice parameters by extending the agent-based model with discrete choice models at finest granularity.