I consulted on such a product a while back and unfortunately the task is much harder than you might think. The project I worked on was actually easier than thi

I consulted on such a product a while back and unfortunately the task is much harder than you might think. The project I worked on was actually easier than this because what we were doing is taking a bunch of the local unsigned bands in Los Angeles (where I live) and trying to use listener preferences to introduce users to those bands. But there were a few significant hurdles... 1. Identification of music is quite expensive. It would be great if a company like Pandora would license their database but until that happens there's a major financial burden on whoever has to pay people to identify existing music in order to make an engine like this work. 2. It's hard to determine user's likes. One of the biggest and most annoying flaws in iTunes is that its "times listened to" count doesn't consider a song listened to until you get to the very end. So if you (like most people) hit "next" when you get to that point at the end of a song where they repeat the lyrics over and over again than iTunes doesn't count that. Which makes it hard for another program to determine your preferences. 3. People have a low tolerance for new music from an artist they don't know. Ideally you want to slip a new song in amongst the listeners favorite songs. But that doesn't work when you have to write your own program (unless you have the money the pay royalties for mainstream songs). Which means you're stuck with trying to make the user listen to 20 new songs in a row which most won't do. Again it isn't a bad idea if someone wanted to try to run with it but I'd consider the above before putting a lot of time into it.