1. Skip to navigation
  2. Skip to content


press room

'Data Mining Music' SXSWi Panel Shows How Data Can Lead to Better Playlists, Recommendations -- With Hilarious Results

March 13, 2012
Billboard
by Glenn Peoples

People have 10 million songs in the pockets: What happens next? Paul Lamere, director of developer platform at the Echo Nest, thinks “big data” is the solution to better listening experiences, recommendations and playlist creation.

“Just as MP3 transformed music in the ‘90s, big data will do the same in this decade,” he told the audience at his SXSW panel, “Data Mining Music.”

Check Out All Our SXSW Coverage Right Here

During an hour-long presentation that had roughly 250 people laughing and clapping, Lamere showed what can be done with music data using two datasets: the 10 million song-strong MusicBraisz and Million Song Dataset.

To demonstrate how music data can lead to better recommendations, Lamere showed a clever use of the Million Song Database to answer the question, “Are dubstep fans or metal fans more passionate?” To gauge fan passion, Lamere looked at the tracks fans actually owned and how often they listen to them. The results elicited some gasps and murmurs: In Flames has the most passionate fans - 115 plays per listener - on a list dominated by heavy metal acts. The Beatles come in second at 95 and Radiohead third with 79. At the bottom was “Eye of the Tiger” performer Survivor with 5 plays per fan, an indication of a one-hit wonder.

A music service makes itself better when it can bring a higher degree of relevance, said Lamere. Last.fm is being too relevant when it suggests obvious artists George Harrison, Wings and Yoko Ono for fans of the Beatles. iTunes can make irrelevant recommendations based on the Beatles when it suggest unrelated and diverse artists who were purchased by buyers of the Beatles. Lamere suggested to instead take into account popularity, familiarity, years of activity, artist relations and fan passion.

Sometimes data can make interesting but fun playlists. Computers are strong at finding paths, and another way to create playlists is to follow a path from one artist to another and listen to every artist along the way. Lamere’s Boil the Frog app creates a single path between any two artists using artist similarity data.

As a demonstration, Lamere created to path from smooth jazz artist Kenny G to death metal act Nile with hilarious results. On the path were Stephen Bishop, Christopher Cross, Michael McDonald, Hall & Oates, Fleetwood Mac, Eric Clapton, Jimi Hendrix, the Mars Volta, the Dillinger Escape Plan and Meshugga. “I consider that playlist to be the highlight of my music-tech career,” Lamere joked.

Billboard.biz is all over SXSW this year—check back several times a day for the latest!

'Data Mining Music' SXSWi Panel Shows How Data Can Lead to Better Playlists, Recommendations -- With Hilarious Results

Press Inquiries

press@matrixpartners.com 408-776-9727