May 18, 2005
Movie recommendations
Yesterday Ben Schafer from UNI gave a talk here on the subject of "recommender systems". Based on the abstract, it sounded a little... fluffy, but I was very pleasantly surprised. He was using some well-grounded AI techniques to process the various preferences of a lot of people and collate them into a single recommendation for you (based on what the system knows you like).
All of which is just an intro to say, go check out this site: MovieLens has you rate 15 movies you've seen, and then gives you recommendations on others. It seems pretty prescient so far.
"Woe unto those who reject love, mercy, compassion and forgiveness
because of what was written in a long-passed time, place and language.
We know they do not know the Living God because they must submit
themselves to a dead thing and give it the name of Jesus Christ."
--Jonathan Prykop
Posted
by blahedo
at 5:28pm
on 18 May 2005
Holy cow, this is addictive!
I gave it a go. Rated about 130 films in less than half an hour.
Most of the predicted ratings seemed accurate, generally within a half a point. Though there were a few amusing exceptions, like Star Trek V, which it predicted I'd give a 3, when I gave it a 1, if that.
Dunno. In the recommended list, I didn't see anything I hadn't yet rated, but really liked. Heck, in the recommended list, I hardly saw anything I previously saw, which was suprising.
It was also recommending a bunch of old movies to me too, like stuff from the 1940's and 50's. I do appreciate movies that actually have a coherent plot, but the old stuff doesn't seem to grab my attention, at least when I'm at the video store.
Okay, that's officially the coolest thing ever. It's actually inspiring us to sign up with Netflix again, because there's a lot we want to see, and even more that we want to re-watch.
I find it harder and harder to determine whether I agree with the predicted ratings, though. It's hard to second guess the website. :-)
