"People who bought ______ _____ also bought..." That gives me almost no new information about the music "they" also bought. It can also be downright disturbing when following this chain of CD's quickly leads back to where you started or to some top 10 compilation. I think it's the "they" of "they also bought" that needs to dominate music selection. Music recommendation engines are essentially 0 dimensional, generally just heaps of similar stuff. True there is that cool site someone mentioned with the sort of web-like connections between bands. But this is essentially leapfrogging from one bin to the next. There is information in those gaps. I think one of the ways to extract it is through time.
Amazon, for one, needs to reorganize recommendations along the lines someone saying "I liked this, then I liked this, and then... I came back and boy do I despise that first thing." This is actually a structure in which you could participate and importantly, easily diverge from onto other tracks. Each person would sculpt a musical landscape whose predictions of what's on the horizon would not have a pixilated feel to them because the musical topology would rely on the global process of music discovery.
Try running any relationship on a kind of pin-hole frame-by-frame basis and it becomes obvious that transmitting meaningful information that way is very difficult. Music choice is about intensely personal relationships with music and music makers. The "I like this or that approach" can be useful, but sometimes it masks other relationships, linked through personal histories, that simply need to be realized. The future is shorter than ever in the digital era. Without this framework it may rush by and all we will have heard is a haunting Doppler wail of wonderful music that's distorted by a freeze frame approach. This music is spot on, yet personally edgy, so that it always opens doors, but never hangs mirrors.
Time is the classic piece that's left out of neuroscience research, which is probably why I brought it up, but I'm sure there are other interesting dimensions through which to organize music preference that could be used to balance preference and diversity in music. What are they?