Haven't signed into your Scholastic account before?
Teachers, not yet a subscriber?
Subscribers receive access to the website and print magazine.
You are being redirecting to Scholastic's authentication page...
Announcements & Tutorials
Renew Now, Pay Later
Sharing Google Activities
2 min.
Setting Up Student View
Exploring Your Issue
Using Text to Speech
Join Our Facebook Group!
1 min.
Subscriber Only Resources
Access this article and hundreds more like it with a subscription to Science World magazine.
DAVE J HOGAN/GETTY IMAGES (SWIFT); KEVIN MAZUR/GETTY IMAGES FOR NARAS (MARS); GLENN HUNT/GETTY IMAGES (SHEERAN); KEVIN MAZUR/ONE LOVE MANCHESTER/GETTY IMAGES (GRANDE); KENTOH/SHUTTERSTOCK.COM (WAVES)
Article Options
Presentation View
Music Master
Oscar Celma helps Pandora analyze your favorite tunes so the app can recommend more songs you’ll love
JOSH MILLER/CNET/YGS GROUP
TASTE TRACKER: Oscar Celma makes sure Pandora’s recommendations match your preferences.
How do you decide what music to listen to? If you’re like the 70 million people who use the music-streaming service Pandora, you rely on their app to help find new tunes that you’ll enjoy. A listener just types in an artist he or she likes—say, Ariana Grande, Taylor Swift, or Drake. Then Pandora uses the information to create a custom radio channel filled with songs by similar artists it thinks will match the person’s taste.
Oscar Celma, the head of research at Pandora, helped design the app. He’s a computer scientist and trained musician. Celma leads a team at Pandora that analyzes data from the app’s users to generate music recommendations. It’s the team’s job to ensure Pandora can predict listeners’ preferences with incredible accuracy. Celma spoke with Science World about how his work combines his love of music and knowledge of computers.
How did your background lead to a job at Pandora?
When I was 12, I started learning to play guitar. Music became my passion. Around the same time, I also started enjoying computer programming—writing instructions, called code, that tell computers what we want them to do. I majored in computer science when I was in college. Afterward, I worked on a Ph.D. project that mixed both my passion for music and computer science. I built algorithms—sets of rules that a computer uses to solve a problem—for recommending music. I tried to improve the music my program suggested by including songs that weren’t as well-known to users. My code was the beginning of what I would eventually go on to create for Pandora.
How does Pandora predict what music a person will like?
PANDORA
SONGS YOU’LL LOVE: The app creates a personalized playlist of new songs and artists based on those you’ve enjoyed in the past.
Based on which music you choose first, the app can guess which other artists and genres you’ll like and make song suggestions. You continue to provide data about your preferences whenever you interact with the app. We call these interactions “signals.” For instance, if you skip or click thumbs-down for a song, that signals that you didn’t enjoy something about it. Or if you want to listen to one song again and again, that signals that you really enjoy something about that music.
We get more than 1 billion signals each day from users. The algorithms take note of these behaviors and can determine patterns in the music you’re listening to. Even if you don’t listen to Drake for a week or so, for example, it knows you like the artist because you’ve listened to him before.
How does your team make sure the app chooses the best songs for users?
I oversee a team of 80 people. Some are data scientists who use computers to make sense of complex digital data. Others are musicologists who study all aspects of music. The musicologists listen to each song Pandora offers, one at a time. They annotate all the attributes of the music, like the key, the rhythm, and the style of the singing. The data scientists integrate this musical data with the signals they receive from users. They’ve built more than 70 algorithms that pull together all this information to find other music that you might like. The recommendations are tailored to you, personalized based on your interests and your interaction with Pandora.
What advice do you have for a young person who’s interested in a job in computer science?
Coding and algorithms have strong components of math and statistics. So programmers need to have a good understanding of these subjects. But instead of focusing too heavily on what you need to know for the future, right now young people should really just find things they enjoy doing. Even playing with LEGO bricks or playing computer games helps a lot in becoming a computer scientist. What starts with LEGOs or games eventually turns into building algorithms. It has to be an evolution, without too much pressure to be the best at math or science. Everyone should be welcome to explore this world.
RELATED CONTENT