Hi Sebastian, What do you think of the technique of retraining the model *only* on the new user, in order to learn the factors associated with him. (Yehuda Koren mentions it here: http://www.netflixprize.com/community/viewtopic.php?id=815)
Is it worse than folding-in? Thanks ________________________________ From: Sebastian Schelter <[email protected]> To: [email protected] Sent: Friday, October 12, 2012 7:00 PM Subject: Re: Recommendations for new users Hi Ahmet, The best way to immediately recommend items to a new user is to fold him into the user space. This is currently not implemented in Mahout, mainly because nobody contributed it yet :) However, this is supported in the ALS implementations of Sean Owen's recommender system called myrrix [1] as well as in an open source weblayer called kornakapi [2] [1] http://myrrix.com/ [2] https://github.com/plista/kornakapi On 12.10.2012 17:31, Ahmet Ylmaz wrote: > Hi, > We are planning to use Mahout for our movie recommender system. And we are > planning to use SVD for model building. > > When a new user comes we will require him/her to rate a certain number of > movies (say 10). > > In > order to recommend movies to this new user we have to rebuild the > entire model. But this not appealing in terms of computational load. > > I'm looking for better solutions. > > For > FunkSVD, one solution seems to be retraining the model *only* on the > new user, in order to learn the factors associated with him. > Since there are not many ratings associated with the new user you can learn > the new user's factors in a quite negligible time. > > Actually > this solution seems not to be difficult to implement. So, I wonder why > this is not implemented in Mahout given that in commercial settings it > is very important to be able to immediately recommend items to users > after they give some ratings. > > Thank you > Ahmet >
