The term of collaborative filtering [Goldberg, 2000] denotes techniques using the known tastes of a group of users to predict the unknown preference of a new user. The distinctive feature of current collaborative filtering processes is to be centralized. The scientific problems have consisted in finding a way to distribute calculus, in order to provide scale for several ten thousands of people, or then to preserve anonymity of users (personal data remain on client side). The impact of a model combining several existing methods to share out tasks between the server and users terminals is examined in this article. A partnership with the company ASTRA allows to turn to their database and to carry out life-sized tests in order to verify efficiency of proposed solutions.