When you have a friendship graph and listen info, you can both match users by their taste and recommend music based on their graph. last.fm did both and more. You can also do non-social personal and non-personal recommendations based on artist "similarities" derived from correlating user tastes. Last.fm does that too.

Another thing is that social network tracks social graph, not "friends" in the narrow sense; some of people in my social graph are likely to listen to the same kind of music than I (active friends); some are likely to listen to white noise (extended family, former close friends, collegues and people I studied with). You basically need circles!