Gaming Netflix

March 6, 2007

I don’t rate movies based on what I think of them. I rate them based on whether I think I would like similar movies. I do this all the time, rating certain mediocre things as good and rating other mediocre things as awful. It basically comes down to this: I don’t trust their ratings algorithm to accurately identify what I like from accurate ratings. Why? Because when I rate an objectively cheesy and dumb movie as good because it has some sort of ideosyncratic value to me, Netflix uses that information to recommend a bunch of terrible movies. And, similarly, when I rate an objectively good movie as bad because it just didn’t quite work, Netflix removes dozens or hundreds of recommendations for things that, in all likelihood, I would enjoy.

Here’s an example. I have been watching Undeclared, Judd Apatow’s TV followup to the ill-fated but excellent Freaks and Geeks. My honest opinion of Undeclared is that it’s just okay. I honestly find listening to the commentary tracks more entertaining than watching the regular shows. But because I like most of Apatow’s other work (Anchorman, 40 Year Old Virgin, etc.), I gave it 4 stars on Netflix, so that it would recommend more Apatow-ish shows/movies to me.

I can’t imagine I’m the only person who tries to anticipate what a recommendation engine will do and aims to game the system to get optimum results.

Obviously, this is relevant for any service that is based on a recommendation engine. I use Amazon’s stars the same way, and there’s lots of other similar stuff going on in other areas.

Bonus content:
Definition of the day: Merriam-Webster’s Online Dictionary defines “monopsony” as “an oligopsony limited to one buyer.” Tremendously helpful, thank you MW.

Posted in Movies, Tech