Yes. That is a good basic recommendation system. Another approach is to use the co-occurrence matrix to find items that have anomalous co-occurrence and then build a weighted model based on overall frequency. This allows you to weight the recommendations differently than you would with the raw co-occurrence score. If you have the right audience and interface then you will still do quite well even with some moderately poor ordering of the recommendations because your viewers will dig pretty far down into the list. Some other interfaces are not so forgiving (think radio).
Variants on that include finding a latent variable representation of movies and people that explains which movies people have seen. The movies you have seen will define a latent variable representation for you and that should allow you to determine which movies you should have seen. This general approach subsumes LSI, pLSI, LDA, MDCA and non-negative matrix factorization for different definitions of latent variable structure. It would be nice to be able to distinguish moves that you have not seen because you never heard of them from movies that you declined to see, but in many domains where marketing is not such a strong effect, you can presume that all things that you have not consumed are things you know nothing about. For movies, this is a weak approximation, for music it is slightly better, for user generated content it is very accurate. On Sat, May 31, 2008 at 8:59 AM, Sean Owen <[EMAIL PROTECTED]> wrote: > Ted coming back to your message here as well -- let me see if I > understand how to apply this. > > Imagine we know which users have seen which movies (no ratings, just > the set of movies). For each movie I know how many users have seen > both, or only one, or neither. I can use this to figure out pairs of > movies that are seen together unusually often. > > Then for each user, find the movies among those they haven't seen > which are most likely to occur together with the ones they have seen, > and recommend them. > >
