Going Head To Head With Genius On Playlists

brownerthanu writes "Engineers at the University of California, San Diego are developing a system to include an ignored sector of music, dubbed the 'long tail,' in music recommendations. It's well known that radio suffers from a popularity bias, where the most popular songs receive an inordinate amount of exposure. In Apple's music recommender system, iTunes' Genius, this bias is magnified. An underground artist will never be recommended in a playlist due to insufficient data. It's an artifact of the popular collaborative filtering recommender algorithm, which Genius is based on. In order to establish a more holistic model of the music world, Luke Barrington and researchers at the Computer Audition Laboratory have created a machine learning system which classifies songs in an automated, Pandora-like, fashion. Instead of using humans to explicitly categorize individual songs, they capture the wisdom of the crowds via a Facebook game, Herd It, and use the data to train statistical models. The machine can then 'listen to,' describe and recommend any song, popular or not. As more people play the game, the machines get smarter. Their experiments show that automatic recommendations work at least as well as Genius for recommending undiscovered music."

Comments

So, not at all?

[BadAnalogyGuy]

It's well known that radio suffers from a popularity bias, where the most popular songs receive an inordinate amount of exposure. In Apple's music recommender system, iTunes' Genius, this bias is magnified. An underground artist will never be recommended in a playlist due to insufficient data. ...
Their experiments show that automatic recommendations work at least as well as Genius for recommending undiscovered music

So, not really so much at all...?

Re:They should go through my collection...

[SlashdotOgre]

Actually I've found last.fm's recommendation system works extremely well; so well in fact that I constantly have a tab open to it when I'm browsing music stores like eMusic (eventually I want to write a little app for this purpose using last.fm's API, but I digress). For those unaware, last.fm users submit what they're listening to through automated plugins (and the supported apps list is huge and very platform independent, I personally use both Amarok 1.4 and MPD); one of the things last.fm does with this music is identifies your "neighbors" (people with similar lastes, i.e. 8 of our top 10 artists are identical). I've found that one of the best ways to find new music is by browsing what my neighbors are listening to and checking out any of their top bands that I'm not familiar with. They also list related artists by correlating this information (e.g. the majority of users who have Band A as a favorite artist also like Band B). Another useful feature is being able to check what an artists most played songs are (great for when it's an artist you never heard of). With that said, I'm definitely interested in seeing what recommendations come from this UCSD team (and not just because I'm an alumnae) as I'm always interested in finding new artists, especially smaller and local ones.