1: \begin{abstract}

2: Collaborative recommendation is an information-filtering technique that

3: attempts to present information items that are likely of interest to an

4: Internet user. Traditionally, collaborative systems deal with

5: situations with two types of variables, users and items. In its most

6: common form, the problem is framed as trying to estimate ratings for

7: items that have not yet been consumed by a user. Despite wide-ranging

8: literature, little is known about the statistical properties of

9: recommendation systems. In fact, no clear probabilistic model even

10: exists which would allow us to precisely describe the mathematical

11: forces driving collaborative filtering. To provide an initial

12: contribution to this, we propose to set out a general sequential

13: stochastic model for collaborative recommendation. We offer an in-depth

14: analysis of the so-called cosine-type nearest neighbor collaborative

15: method, which is one of the most widely used algorithms in

16: collaborative filtering, and analyze its asymptotic performance as the

17: number of users grows. We establish consistency of the procedure under

18: mild assumptions on the model. Rates of convergence and examples are

19: also provided.

20: \end{abstract}