Back when the net was younger, there was this cool system called RINGO. You'd send an email with 10 albums and what you thought of each one. Sometime later you'd receive a list of things you might like along with a prediction of how much they think you'd like each one. Sounds familiar, right? Sure Amazon has it and allmusic.com has it, but for some reason RINGO always came up with a great selection. I got turned on to a bunch of stuff I would of never found otherwise.
I'm not sure why it worked so much better than the systems we have today: was it because of the people using it or was it just a better system?
Perhaps we'll never know, because I can't seem to find the source code. I know the developers at MIT changed the name to Firefly then either sold it or formed a company around it. Then I think it got absorbed into some website or another. Then they added movie recommendations and book recommendations, but neither of those worked very good. Then it seemed like the music stuff started working less good and then I seem to recall it became a fee-based system.
So I would love to get ahold of the original version!
When they migrated it to the web, the page was http://rg.media.mit.edu/ringo/confirm.html to confirm your subscription
In 2001, I found a few papers describing the system floating around the web. The 93 page dissertation by Upendra Shardanand from September 1994, called Social Information Filtering for Music Recommendation. Then, by the same author, a paper from 1995 describing some internals called Algorithms for Automating ``Word of Mouth'' in ascii markup, pdf and postscript.
It looks like Paul Perry is doing some work with this in sql: A Minimalist Implementation of Collaborative Filtering
If anybody has leads of where to find more info or that old source code, please let me know!
An interesting implementation of a similar idea has been done as a liturature recommendation system called AlexLit