Probabilistic models for data combination in recommender systems
Sinead A. Williamson, Zoubin Ghahramani · 2008
In a typical collaborative filtering problem, the dataset is an incomplete matrix of ratings R given by a set U of users to a set I of items, and the task is to predict what ratings the users would give to the items they have not yet rated. A common approach to this problem is to use matrix factorization techniques to find a lower dimensional representation, R ≈ UM T