I just joined Netflix and have noticed a feature they have on there to automatically recommend films for you based on your ratings of films you’ve seen. It’s a feature similar to Amazon.com’s “You may also be interested in” recommendations, only Amazon seems to look at customers buying more than one item and correlates them, whereas the Netflix feature seems to adjust the ratings it thinks you’ll give a film based on ratings you’ve given another film. I don’t know what complex algorithm they use to come up with these ratings - the FAQ is mute on this point - but it’s clearly not complex enough: when I rated more films it seemed to become less accurate in predicting my tastes. Maybe my film tastes are too complicated (not very likely) or it’s just impossible to predict people’s tastes from their five-star ratings. But I sense something else is behind it; the algorithm, whatever it is, seems to be based on categories of films (Classics, Independents, Foreign, etc.) rather than on categories of aesthetic perception that might drive film selection - in my case modernist-auteurism, camp sensibility, and lack of overinvestment in “independent” or “foreign” as a marker of quality. The trick, of course, would be to sketch out major modes of film appreciation so that the algorithm could then put the customer in the right category, for which aesthetic decisions would have more context. The least helpful thing about these rating systems, including Amazon’s, is that you have action fans giving a Dreyer film one star because it’s boring and auteurists giving action films one star because the dialogue is poorly written, so neither group gets useful reviews or ratings. Of course, a more accurate system would involve sociological research beyond our needs, when we all can usually find films we want to watch on our own, or from friends’ recommendations. Still, it’s an interesting case study that calculating variables will produce the wrong result if the wrong variables are used to begin with, a lesson often lost on marketing and public opinion research.
No comments have been added to this post yet.