Time to plow through my backlog of stuff to post. Last month I went to the BayCHI meeting on Beyond Search: Social and Personal Ways of Finding Information, which was about social search and recommender systems. I got a few things out of that meeting.
How do you recommend to someone that they'll like obscure song B if they like popular song A? How does the connection between those two songs get made in the first place? The speakers from Netflix, Live365, and Pandora agreed that you need experts to make that connection, since the public at large doesn't know enough to make that connection on its own.
Also, if a person likes a popular song, you can't recommend a completely obscure song that appears unrelated even if it is. The user must be gently led down the long tail.
Finally, I started using Pandora for listening to music. I gave it one song, "Venus de Milo" by Miles Davis, and it set up a station that played that song and about 100 other songs that are similar. An instant instrumental jazz station, perfect for listening to at work.