Pandora, the Internet radio service, is plying a new tune.
After years of customizing playlists to individual listeners by analyzing components of the songs they like, then playing them the tracks with similar traits, the company has started data-mining users’ musical tastes for clues about the kinds of ads most likely to engage them.
“It’s becoming quite apparent to us that the world of playing the perfect music to people and the world of playing perfect advertising to them are strikingly similar,” said Eric Bieschke, Pandora’s chief scientist.
Consider someone who’s in an adventurous musical mood on a weekend afternoon, he said. One hypothesis is that this listener may be more likely to click on an ad for, say, adventure travel in Costa Rica than a person in an office on a Monday morning listening to familiar tunes. And that person at the office, Mr. Bieschke said, may be more inclined to respond to a more conservative travel ad for, say, a restaurant-and-museum tour of Paris. Pandora is now testing hypotheses such as these by, among other methods, measuring the frequency of ad clicks. “There are a lot of interesting things we can do on the music side that bridge the way to advertising,” said Mr. Bieschke, who led the development of Pandora’s music recommendation engine.
A few services, such as Pandora, Amazon, and Netflix, were early in developing algorithms to recommend products based on an individual customer’s preferences or those of people with similar profiles. Now, some companies are trying to differentiate themselves by using their proprietary data sets to make deeper inferences about individuals and try to influence their behavior.
This online ad-customization technique is known as behavioral targeting, but Pandora adds a music layer. Pandora has collected song preference and other details about more than 200 million registered users, and those people have expressed their song likes and dislikes by pressing the site’s thumbs-up and thumbs-down buttons more than 35 billion times. Because Pandora needs to understand the type of device a listener is using in order to deliver songs in a playable format, its system also knows whether people are tuning in from their cars, from iPhones or Android phones, or from desktops.
So it seems only logical for the company to start seeking correlations between users’ listening habits and the kinds of ads they might be most receptive to.
“The advantage of using our own in-house data is that we have it down to the individual level, to the specific person who is using Pandora,” Mr. Bieschke said. “We take all of these signals and look at correlations that lead us to come up with magical insights about somebody.”
“I would guess, looking at music choices, you could probably predict with high accuracy a person’s worldview,” said Vitaly Shmatikov, an associate professor of computer science at the University of Texas at Austin, where he studies computer security and privacy. “You might be able to predict people’s stance on issues like gun control or the environment because there are bands and music tracks that do express strong positions.”