Here is one reference which I have used earlier for a class project: http://jmlr.csail.mit.edu/papers/volume6/shani05a/shani05a.pdf
One can think of recommendation as an instance of sequential optimal decision making problem, something which MDPs have been traditionally used for. Given the distributions, solving an MDP boils down to arriving at the Policy table enlisting optimal actions to be taken for each state. One can use a couple of different algorithms--Policy Iteration or Value Iteration to solve it. If no prior distributions are given, it becomes a more Machine Learning style of problem. Q-Learning is generally employed in those cases. On Sat, Oct 8, 2011 at 6:04 AM, Grant Ingersoll <gsing...@apache.org> wrote: > I haven't seen any discussion on it. Do you have a paper or other > reference on it? That usually helps in discussing how to go about it. We > have HMM for classification. > > On Oct 7, 2011, at 10:33 PM, Colin wrote: > > > I haven't find any MC-based open source recommender. > > Does Mahout have any plan to provide some? > > > > Thank you, > > -- > > Colin Wang > > Skype : colin.bin.wang > > -------------------------------------------- > Grant Ingersoll > http://www.lucidimagination.com > Lucene Eurocon 2011: http://www.lucene-eurocon.com > >
