College of Engineering, Computing and Applied Sciences

From Drake to disco: Daniel High student teams up with Clemson University to study diversity in music streaming services


The rapper Drake, disco music and three researchers with Clemson University ties are among the factors that have helped put Ian Baracskay on track to become a published researcher while still a senior at D.W. Daniel High School.

Baracskay’s work on music streaming services started as a project for an AP Research course and later evolved into a paper that has been accepted to the Association for Computing Machinery’s 27th Annual Conference on Intelligent User Interfaces.

Bart Knijnenburg, associate professor of human-centered computing, works in his McAdams Hall office.

The paper is titled “The Diversity of Recommender Systems” and explores whether the variety of music that popular streaming services recommend to their users is correlated to the services’ success.

“Technically, the paper isn’t published yet,” Baracskay said in an interview before the conference. “But it does feel pretty neat to have the opportunity to have already published something when most students don’t get to do it until they are graduate students or undergraduates.”

Bart Knijnenburg, associate professor of human-centered computing at Clemson, served as Baracskay’s advisor on the project and looped in Ph.D. student Mehtab “Shahan” Iqbal to help advise the team.

When Baracskay needed help with coding, he turned to Donald Baracskay III, who received a Bachelor of Science in computer engineering from Clemson in 2018 and is now studying artificial intelligence as a master’s student at Northwestern University. He also just so happens to be Ian’s older brother.

Knijnenburg said the group submitted the paper as “late-breaking” work. The category acts as a showcase for clever, potentially inspiring ideas that are in the early stages of research.

“I thought this was exactly that kind of paper,” Knijnenburg said. “We can look at the recommendations these algorithms give and how diverse the recommendations are and then start thinking about what that means for the end user.”

Ian Baracskay

Ian said that when he first chose his topic, his AP Research teacher told the class to find an expert advisor. Ian went to the Clemson website that lists the computer science faculty.

The first professor he emailed referred him to Knijnenburg, who has done extensive research on recommender systems, including a project funded by a CAREER award from the National Science Foundation.

“This was right in my wheelhouse,” Knijnenburg said. “If you have a study on recommender systems that is not focused purely on the algorithm but specifically on how people perceive recommendations, you can email me something like that any time of day because that’s what I do.”

For its research, the group created three playlists of 20 songs each. A low-diversity playlist contained only disco from the late 1970s and early 1980s, a medium-diversity playlist featured rock music from 10 different artists and a high-diversity playlist contained a wide range of genres from 20 different artists varying from Drake to Liszt.

Those playlists were loaded into five different streaming services with separate accounts for each playlist to prevent “recommendation bleed-over.” Researchers took note of the first 30 songs that the services’ algorithms suggested.

Then they used a matrix factorization algorithm trained on the Spotify Million Playlist Challenge Dataset to calculate a similarity coefficient for each set of recommendations. The lower the similarity coefficient, the higher the diversity.

Mehtab “Shahan” Iqbal talks with Bart Knijnenburg while sitting next to Sushmita Khan.

YouTube Music had the lowest similarity coefficient, followed by Apple,, Pandora and Spotify.

“We found that YouTube Music offered the most diverse recommendations, but the perception of the recommenders was similar across the five services,” the researchers wrote. “Consumers had multiple perspectives on the recommendations provided by their music service– ranging from not wanting any recommendations to applauding the algorithm for helping them find new music.”

Researchers also found that most services tend to match the level of diversity of the playlist. So a more diverse playlist usually results in more diverse recommendations.

“In effect, each service is able to cater to users’ preferred level of diversity, as expressed in the songs they listen to organically,” researchers wrote.

The work on the paper gave Iqbal a chance to strengthen his hand at teaching and mentoring, areas of increasing interest to him since he began studying under Knijnenburg. Iqbal said he liked that Ian was curious and that he was impressed with the questions Ian asked.

When Ian expressed doubts about not having a university affiliation to attach to his name, Iqbal and Knijnenburg were there to reassure him.

“Do you realize that’s the coolest thing?” Iqbal remembered saying in a text to Ian. “You’re going to be in a venue with universities, professors and Ph.D. students from around the world, and you’ll be in high school. Someday you’re going to look back at this. You have no idea how cool that is.”

The team summarized its findings on this poster.

Amy Apon, the C. Tycho Howle Director of the School of Computing, said that Knijnenburg did an excellent job of turning an email from a local high school student he didn’t previously know into a public service project with multiple benefits.

“This project gave Ian an exemplary research experience, Shahan a chance to practice mentoring and Don an opportunity to connect with his alma mater– and his younger brother,” she said. “Further, this research provides new ideas for studying recommender systems. I congratulate the team, especially Ian, on its success.”

Ian is first author on the paper. He wrote his AP Research paper on his own, and Knijnenburg and Iqbal helped him reframe it for the conference, which is set for March 22-25.

The conference was originally supposed to be held in Helsinki, Finland but has since gone virtual. Knijnenburg serves as the conference’s doctoral symposium co-chair.

The Baracskays are the sons of Karen Baracskay, a biology instructor at Tri-County Technical College, and Donald Baracskay II, a psychiatrist.

Donald Baracskay III said that after he gets his master’s degree, he will pursue a job in artificial intelligence and may go for a Ph.D. later.

His younger brother said that he has applied to several colleges to study computer engineering or computer science and is still deciding where he would like to attend.

But Clemson is the only university he would be able to reach by bicycle from where he lives now.

“I think he has a bright future ahead,” Knijnenburg said. “I already told him that if he comes to Clemson, he can work with me. But if he goes to any of the other schools, he can work with me, too, if he wants. I always say people with unique ideas are the best people to work with, especially if they have brillant siblings they can ask for help once in a while.”

Want to Discuss?

Get in touch and we will connect you with the author or another expert.

Or email us at

    This form is protected by reCAPTCHA and the Google Privacy Policy and Terms of Service apply.