Pandora or last.fm?

I've been a last.fm fan for quite a while now, but today I discovered Pandora, which might turn me into a defector. The two services work quite differently. last.fm assumes that if you enjoy music popular with a group of listeners, you'll enjoy other music popular with those same listeners. The last.fm recommendation algorithm works solely on the basis of users' thumbs up/thumbs down reports, knowing nothing about the actual qualities of the music itself. Compare this with Pandora, which bases its recommendations on the work of a team of music analysts, who have classified songs according to many properties (melody, harmony, instrumentation, rhythm, vocals, lyrics, etc), and recommends songs on the basis of these properties. It's early days yet, but based on my experience so far, Pandora's recommendations seem superior. If you're interested, I've included a link to my Pandora favorites in the sidebar of this blog. You can listen to my station by clicking on cyberchaos radio. Although I'm pleased with Pandora, I can see that it has a dangerous bottleneck -- the system can only introduce new songs as quickly as a team of human experts can analyze them, whereas a system like last.fm can add new songs to its database and let users start rating them immediately.

Tags: music recommendation | pandora | last.fm | collaborative filtering | recommendation engine