Ken's Voyages Around the Sun

The Netflix Prize
Previous Entry :: Next Entry

Read/Post Comments (4)
Share on Facebook
Our online DVD rental company, Netflix, has announced The Netflix Prize. Inspired by the 18th-Century Brtish Longitude Prize, this new one will award a million bucks to whomever can develop slightly better movie-matching software for their customers. You need only improve on it by 10% to win.

It's mighty tempting to sign up for the contest, but sheesh, I have a hard time believing in the premise and given the "mere" 10% improvement they want, no doubt it's rather hard to accomplish. They're giving people five years to work on it.

When I say it's hard to believe in the premise, it's because Netflix has people rate movies on a simple 1-5 star scale. Such a generic feedback system seems wholly inadequate to predict what other movies you'll like based on the ones you have rated even if you have thousands as we do.

There are so many things to consider when rating a movie that I don't see it happening with such a simple rating system.

How accurate are the ratings that people make in the first place? That is, like in our household, who rates the movies and how carefully are they rated and are the ratings consistent?

Would you rate a given movie the same depending on whether
you saw it before or after another one? (Impossible to know, of course, but I would expect it to be a factor.)

Do you go back and re-rate your movies periodically? I doubt people do this but they may if they truly hope Netflix's recommendations system will deliver them what they want to watch.

Do you have your own little rating system that takes into account the various aspects of the movie such as directing, acting, production values, historical accuracy, soundtrack, plot, etc., etc.? If so, how do you weight each aspect? Do you translate your score to the one-to-five star ranking that Netflix gives?

And how do you account for people who may like to watch from a large number of genres but also other things that friends recommend or that win prizes or might be rented because visiting family members want to watch them?

In short it just seems there are too many variables and unknowns and unknowables used with a too-imprecise measurement system to offer the improvement they want.

You software engineers out there... how say you?

Read/Post Comments (4)

Previous Entry :: Next Entry

Back to Top

Powered by JournalScape © 2001-2010 All rights reserved.
All content rights reserved by the author.