Breese et al. [15] implement a Bayesian Network where each node corresponds to each item. The states correspond to each possible vote value. In the network, each item will have a set of parent items that are its best predictors. The conditional prob- ability tables are represented by decision trees. The authors report better results for this model than for several nearest-neighbors implementations over several datasets. Hierarchical Bayesian Networks have also been used in several settings as a way to add domain-knowledge for information filtering [78]. One of the issues with hierarchical Bayesian networks, however, is that it is very expensive to learn and update the model when there are many users in it. Zhang and Koren [79] propose a variation over the standard Expectation-Maximization (EM) model in order to speed up this process in the scenario of a content-based RS.