Probabilistic models for data combination in recommender systems

Collaborative filtering systems suffer from two significant, and related, problems: sparsity and extension to new items. Due to the fact that most users rate a very small subset of the universe of items, the user-item rating matrix is generally very sparse, particularly in the case of new users or obscure items, leading to poor predictions in such cases. Furthermore, an item cannot be recommended until it has been rated, which is a significant disadvantage since most commercial recommender systems will hope to promote new items.