|
I need to combine popularity rating for movies with predicted rating(s). What is the best way to do this? Ideally I'd like to use the same method to combine rating predictions coming from different experts and popularity (which could be seen as another expert) using the same method. I see Armitrain's presentation on the subject http://i.stanford.edu/infoseminar/amatriain.pdf He talks about using linear regression to fit a line btw popularity / personal ratings on projecting onto that line. How could this method be extended into multiple ratings from different experts? Amatriain talks about Learning to Rank literature - in this area he mentions RankSVM, etc. I did some reading on the subject, and one idea I have is using logistic regression - all predicted ratings from all experts and popularity would do into LR as dimensions, label would be relevant / non-relevant (we have two ratings in our system), and LR is trained against that. Then, after rating prediction, we feed these into LR, and get another set of probabilities which we use to rank. Would this be okay? |
