Recommendation Systems

I've written a recommendation system for music, movies, games, authors, and politicians (yeah, I know it's strange, but I figure people with similar taste in politics will have similar taste in media). It's in the vein of the systems used by Amazon and Netflix, but with a very different core algorithm, which I believe performs better.

The system works by comparing users' ratings against each other to create a compatibility score. Users who love and hate the same things get a higher score, while users who disagree receive lower ones. This score is applied to all the ratings for a particular item and an aggregate score is created. The items with the highest aggregate scores are suggested to the user. The advantages of this type of system over others (collaborative filtering, et. al.) is that it more closely correlates users' ratings together, is difficult to subvert, and still performs well enough to be used in realtime.

If you're interested in such systems, or are just interested in getting some recommendations, here are a few of my favorite links:

http://www.kindakarma.com/ - music, movies, games, authors, politics (my site)
http://upto11.net/ - music
http://audioscrobbler.com/ - music (this is a plugin that communicates with a central server)
http://moonranker.com/ - music, movies books
http://amphetarate.com/ - RSS feeds (most interesting use of this technology)
http://storycode.com/ - books
http://stumbleupon.com/ - websites (requires a toolbar in your browser, but well worth the effort)