Wednesday, March 05, 2008

Bias and Netflix Prize

The Netflix prize aims to get an algorithm that judges better how people rate items.This article in wired talks about the use of examining human biases to help you rate items.

I find human cognitive biases fascinating and used them to develop a lottery number generator. What biases could affect the netflix prize?

Genetic Bias. As this article describes numerous of our biases/opinions have a genetic basis.

The article reports on how your opinions on X-Rated movies has a genetic component. Now unless you have an identical twin in your Netflix set (they are 0.2% of the world population) this information may not be of much use. Still I think it is interesting that because some opinions have a genetic component peoples Netflix ratings probably have a genetic component.

There is a list of cognitive biases here. If I think of any others that are relevant to the Netflix prize I will post an article.

Off hand I would expect these effects
1.Bandwagon effect. If everyone you know says a film is good you will probably go along with them to some extent.

2. Post-purchase rationalization. Your ratings are likely to be slightly higher then your real feelings. The fact you went and rented the film means you have a certain investment in it being good. If you then say you were wrong you are admitting being wrong to some extent. People hate to admit being wrong.

3.Zero-risk bias. This is tenuous but i think users are more likely to rate a .5 movie as 0 then a 4.5 movie as 4. One bias i have not heard mentioned is how people will order films based on how this will make them look.

This slate article describes he phenomena. I expect that people will also overrate pretentious movies on the grounds that no one want to give Battleship Potemkin as 1 and Porkies 2 as 5.

No comments: