[My first post using Movable Type 3.2]
For many years, various sites on the Internet have offered the promise of analyzing or sampling our musical tastes and intelligently giving us back a playlist of similar music (or at least music you might like...) Along the way we've seen a number of sites come and go with various degrees of success:
Plenty of research has been done and many more students continue to expand the idea pool on how to best correlate and recommend music based on some known set of variables, meta data, or previously established social preferences.
I recently came across Pandora.com, a creation of the Music Genome Project, and was quite pleased with what I experienced. First, this is a pay site ($36/year or $4/month) with a free 10-hour trial period. Pandora starts by allowing you to specify and artist or song, then it returns a "station" that plays music related to your choice. It has a clean, intuitive Flash interface that shows your stations (you can create 10 with the demo, 100 with a subscription.) and displays the last 3 songs played (along with the artist's name and a thumbnail of the album cover.)
You can only pause or skip tracks (no rewinding or replaying on demand) because of the way Pandora structured its licensing arrangements. You are clearly paying a small fee to listen to music intelligently correlated to your tastes, not buying it. If you do find a track you like, you can purchase it through easy links back to Amazon or iTunes. You can also provide feedback (thumbs up/down) on whether or not you liked a parituclar track. The music is streamed to you in an unspecified format at 128kbps. There is no advertising on the site, and they don't censor the music; unlike MTV, they play the music as the artist wrote and performed it. <sarcasm>What a neat idea.</sarcasm>
I enjoyed exploring Pandora's musicscape and found several new artist through its suggestions. For the most part, I found the recommendations very relevant. I tried a variety of artists and the only confusing results were for a channel based on Sublime, where Pandora picked up on the guitar-and-vocal ballads by Sublime instead of the punkier, ska- and reggae-influenced side of Sublime. An accurate match, but not the attribues I was looking for when I typed in Sublime. "Pink Floyd" turned up some psychedelic/prog rock stuff I hadn't heard before. "Mark Knopfler" came back with some interesting folksy balladeers I hadn't heard of before. I ran through a variety of artists including Aimee Mann, Neko Case, The Police, Rush, Goldfinger, Johnny Cash, Kate Bush, Colin Hay, and Primus and got good results for all. The coverage (they claim 10,000 artists and 300,000 songs) seems quite complete and the recommendations are truly similar music, not stuff pushed from the top-40 or top-100 lists.
Pandora accepts submissions from artists as well as suggestions from listeners as well. Check out Pandora.com for the 10-hour demo and see how you like the service. I am not affiliated with Pandora in any way, just a subscriber.
[ Update (February 2, 2006) :
There is a great comparison of Pandora and Last.Fm including an analysis of their algorithms and effectiveness.
Tags: social networking networks research experimental music rating recommend recommendation firefly ringo mit media lab musicstrands last.fm automatic automata automation autonomous algorithms personal personalized community sharing shareable
