Markov Random Fields for Collaborative Filtering

In this regime (n > m), learning a sparse model may not lead to significant improvements inprediction accuracy(compared toadensemodel).