After touring with rock bands for almost a decade, Tim Westergren decided something had to change. The Stanford-trained saxophonist thought driving thousands of miles in an old van was perhaps not the most efficient way of matching independent musicians with potential audiences. His solution to the problem? The Music Genome Project. In 2000, Westergren and a group of collaborators set about analyzing music strictly on musicological terms (e.g., rhythm, pitch, harmony, instrumentation) in order to find the "connective tissue" that would allow for reliable recommendations. Westergren's approach is different from that of Last.fm, MusicStrands or Amazon.com, where recommendations are based on consumption patterns; i.e., what other people are already listening to. Users can access the Genome through a site called Pandora, where, in theory, they'll be able to find an answer to the question, "What kind of music might I like?"
What are the advantages to Pandora's method of recommendation?
Pandora is based on three things: We want it to be easy to use and be on-target quickly. The third thing is that it has to be full of music you've never heard before. Once people get past their 20s, they often don't have the time to find new music. We pointed our company at that problem.
Because we're not about sharing playlists that other people have compiled, we're better at recommending stuff you haven't heard before. We're trying to solve the problem of bands that don't have a playlist history. Some of those other services that make recommendations based on playlist popularity present a Catch-22 -- you have to be popular to be popular.
Just because two artists may share particular musical attributes doesn't mean they will appeal to the same listeners. Is there a hole in Pandora's approach?
There is definitely a socio-cultural or second-order perception that we're not a part of. And I'm kind of glad about that. Sometimes I'll get e-mails from people wanting to know why Neil Diamond is showing up on their acoustic rock playlist and I'll ask, "Does it fit in strictly musical terms?" Often it does, and that person will acknowledge they actually like Neil Diamond.
Music is not served by snobbery and the Genome challenges people's notions of what is good and what isn't. But if enough people were to give Neil Diamond a thumbs down on their playlists when it showed up in a given context, then we would admit that in that instance we got it wrong.
What do you think of the idea that record companies would be interested in having their artists recommended over other similar artists?
We'll never do it. We will never, ever play a song that was put there because it was paid for. Pandora's position is built on trust and if we did that that trust would be lost. That's what's happened with radio. We'd consider having labels advertise with us in a similar manner as Google -- with sponsored links -- but then you know that they're paying for that. I've had largely positive experiences with labels -- the people there are usually working there because they wanted to be around music and lots of time they're pumped about Pandora. Certainly some people have approached me about paying for recommendations, but they drop it once I tell them that we won't do that.
-- David Marchese