In Restricted Boltzmann Machines for Collaborative Filtering Restricted Boltzmann Machines (RBMs) are applied to the Netflix Prize data set.
An obvious next step might be to use stacks of RBMs (i.e. Deep Belief Nets) on the same problem, and indeed this is discussed briefly in section 8.2 of the RBM paper. Has this ever been tried? I'm suspecting that the data in question is not amenable to further levels of abstraction, but I would be interested in others thoughts on that.