There are basically two approaches that I have seen, which I'll label AMAZON and PANDORA for ease of use. In the first you collect user data and suggest what other people with similar "profiles" selected. This is problematic in Amazon where you might be buying gifts that don't reflect your taste, but in Netflix it should be fairly robust. The other is much more labor-intense, and involves semantic coding of the artwork. In the case of Pandora, they invested hundreds of thousands of man hours listening to music and creating a 'genome' for each work, which allowed them to objectively correlate the songs in n-dimensional factor analysis.
Oops, two more approaches: There is also the flickr / del.icio.us use of tags, a FOLKSONOMY, and I think currently Netflix is using a fourth hybrid system which I'll call GOOGLE because they seem to use some database (say, IMDB) that lists directors and they just harvest the relevant links. Hell, maybe Netflix even uses Google to find the top 100 pages that come up for "Kill Bill" and then they just compare all the words in all those pages with all the titles of movies in their database, assuming that movies that are mentioned on the same page are in some way relevant to one another.
But these three options assume that "better" is already defined. Is that an objective property of a suggestion? Its hard to imagine what that would mean. Instead, is it most user-friendly? Which opens the whole FORM-FUNCTION debate. I mean, maybe just a better GUI ("better" would be defined as "more user selections"=more clicks) would satisfy that requirement.
And then there is the question of why they'd really want a better suggester. Their profits are not linked to the number of movies someone rents. So is the million just self-promotion? Or philanthropy to the unemployed programmers of the world?
Now that Google has Youtube, everyone is wondering what they're going to do with it. How do you make money off of indexing a huge database of free movies? Two ways, either LEECH (ie targetted advertizements) or CHARGE (micropayments for content). Micropayments haven't caught on yet, but with the continuing expansion of Google content (ie one login for calendar, email, personalized google search, etc) similar to Yahoo, it shouldn't be too hard to set up micropayment accounts.
Paying for content isn't all bad; according to economic theory, value equals money, and vice versa. If you pay money for something you create valuable information. If it cost a small fee to link another website from your site, the average quality of links would go way up. Similarly, if the post office raised the bulk (junk) mailing rate from 13 cents closer to first class (39 cents) the average amount of bulk would decline, but both because of decreased information overload and because of the recognition of the cost inherent in the medium, people would pay alto more attention to the bulk mail (currently 50% is thrown away unopened). So this would not necessarily be bad for the advertisers. Perhaps the same logical structure could entail for micropayments on media content, although I can't quite see how right now. http://econ.arizona.edu/downloads/Seminars/AndersonF06.pdf
Anyway, information markets are still the best way to self-organize information. More later.