A long time ago a similar service existed, "www.movierater.com", you would rate a mess of movies and it would give you the picks out on video and in theaters that you would likely like. I miss that one, it had pretty good accuracy.
The only thing I don't like about this one is that it is Yahoo doing it. They have too much presence and too much potential for data mining that the old single site I used to use didn't have.
Instead of 'rating' they should offer you two movies and ask 'which do you prefer'. This is a much better way to do it -- you'll notice that all the movies average out to a 'B' rating. It's extremely difficult, for me at least, to rate movies on a simple absolute scale. I tried the Yahoo one, but then realized that I was just rating the vast majority of movies "decently good". Yet I can almost always tell you which of two movies I prefer.
I prefer the A or B approach too, but apparantly it takes longer to build up the baysian model using that approach. I don't know the details though. That is all the science team.;)
A Bayesian inference model would not only take a while to train, it's only good for small populations of users. As soon as you have conflicting expectations, bayes goes out the window. It would be really prohibitive in CPU for yahoo to run such an inference on every query... though for a browser extension, that would be quite doable and nifty.
Actually, we use a method we call Attributized Bayesian Choice Modeling, and it works quite well.:)
Roughly speaking, we classify the movies based on a variety of criteria, from actual attributes such as ratings, directors, actors and such to latent attributes such as whether the movies is thought provoking or thrilling.
We then create a profile of the user's preferences for those attributes.
Finally, we recommend items matching either the preferences that the user has specifically stated *or* items with a
"You must have an IQ of at least half a million." -- Popeye
yahoo! Next (Score:2, Interesting)
http://next.yahoo.com/ [yahoo.com] this one is cool http://farechase.yahoo.com/ [yahoo.com] try this http://movies.yahoo.com/recommendations/ [yahoo.com]
Re:yahoo! Next (Score:2)
Re:yahoo! Next (Score:2)
The only thing I don't like about this one is that it is Yahoo doing it. They have too much presence and too much potential for data mining that the old single site I used to use didn't have.
Re:yahoo! Next (Score:3, Insightful)
Re:yahoo! Next (Score:2)
Re:yahoo! Next (Score:2)
Re:yahoo! Next (Score:2)
Roughly speaking, we classify the movies based on a variety of criteria, from actual attributes such as ratings, directors, actors and such to latent attributes such as whether the movies is thought provoking or thrilling.
We then create a profile of the user's preferences for those attributes.
Finally, we recommend items matching either the preferences that the user has specifically stated *or* items with a