The Art & Science of Music Recommendation Engines

I don’t know about you, but if I ever have to listen to that same, 12-song playlist on the local classic rock station ever again, I might go completely insane. Thankfully, we live in the modern age of music recommendation engines, where I can simply ask for a stream of music that lies somewhere between Wilco and Weezer and, voila!

We spoke with Pandora founder Tim Westergren and Slacker Radio VP of marketing Jonathan Sasse to find out a little bit more about what goes into this seemingly effortless and unending flow of personally catered music.

Sponsor

The State of Recommendation Engines is a sponsored content series by BT Buckets, a leader in personalization and onsite behavioral targeting. Check out their solutions.

Both Slacker Radio and Pandora work to do the same thing – provide users with an unending stream of music catered to their tastes in order to help them discover and listen to music they like – but they go about it in different ways. One is a bit more of a science and the other, a bit more of an art, thought there’s a science and art to each.

Pandora & Musical DNA

Pandora has a catalog of more than 750,000 songs – far fewer than you might see with streaming music services like MOG, Napster or Rdio. Each song is reviewed by a member of Pandora’s review team and characterized according to nearly 400 musical attributes before being added to the Music Genome Project. The project is “the underlying backbone” behind Pandora, Westergren told us.

But how do we get from this to the songs being streamed over your headphones? According to Westergren, the Music Genome Project provides a sort of “musical DNA,” and from there the site is constantly trying to shape what it plays next according to your tastes. On Pandora, users can provide feedback in the form of thumbs-up and thumbs-down to influence the songs the site plays.

“Over time, we try to figure out the patterns in your thumbing,” said Westergren. “It’s really trying to replicate the knowledgeable record store guy – the human who knows what you like and don’t like.”

Giving a thumbs-down to a Joni Mitchell tune, then, won’t limit your playlist to male vocalists, but if you continue over time to thumbs-down female vocalists, the system is designed to pick up on this pattern and cater your channel accordingly.

Slacker Radio – Let Someone Else Do It For You

Slacker Radio takes a decidedly different approach to music recommendation with its catalog of more than 3.5 million songs. According to Jonathan Sasse, a senior VP of marketing at Slacker Radio, the biggest component of the service is its “hand programmed” genre channels. These channels are created by people with “extensive radio programming experience,” Sasse told us.

According to Sasse, Slacker takes a much more hands-on, curation-based approach to music recommendation. Its stations, rather than being created on-the-fly according to algorithmically comparing musical characteristics, are programmed much like radio would be, on and hour-by-hour basis.

“[DJs] have specific control over the exact songs and what the rotation looks like,” said Sasse. “As the community interacts with those stations, we start to see how those songs are being received, whether they’re getting a disproportionate amount of likes or dislikes, and it starts to affect the DJs.”

Art or Science?

In the end, while Pandora appears to take a more scientific approach with its genome project and Slacker Radio looks to keep the art of the DJ alive, neither appears to be simply one or the other and both face the same problem – creating a custom stream of music catered to the individual.

“One big challenge is if you have extremely eclectic taste,” said Sasse. “If you take a collection of artists that have no ties whatsoever, it can be challenging to decide what to play for you at the right time.”

“It’s a really tricky problem and it’s wickedly hard to make it work,” said Westergren. “Tastes are so individual, it really takes a combination of different methods together.”

In the end, both services come down to a common element – the wisdom of the crowd. As more and more data is collected from users, the services become increasingly refined.

Sasse said that, for Slacker Radio, the job of the DJs is to bring emerging content to the top, but the “next step is for the community as a whole to drive that content.”

For Pandora, Westergren noted the more than 6 billion pieces of “thumb feedback” that he says gives the service “a huge amount of insight” that he says isn’t general, but rather contextual.

So, if your well of music recommending friends has finally run dry, fear not because there’s an army of trained DJs and music-categorizing musicians ready to recommend that next band to rock your socks off. And if you don’t like it, you can just give it a thumbs-down and see what’s next.

Discuss