This seminar talk is about using regularized regression to determine the number of latent factor in a recommender system. Also at the last I'll talk about how it is been applied in a real project and how to deal with the significant cold start problem.