EECS PhD student's AI model to predict Eurovision result
Kasia Adamska, a PhD student at the School of Electronic Engineering and Computer Science, has developed an AI model that she says can accurately predict which songs will reach the Eurovision final this weekend and ultimately win the contest.
Kasia has spent two years refining her machine learning model as part of her research project called Predicting Hit Songs: A Multimodal and Data-Driven Approach. She has already predicted which songs will win the semi-finals and will use her model again on the eve of the grand final, to foretell who the overall winner will be.
Kasia’s approach uses data to understand what might contribute to a song’s success in the charts, or in a particular context like Eurovision. Her AI model looks at the songs themselves, analysing features like rhythm, tone, and the repetitiveness and sentiment of the lyrics. It also takes in the wider context, like the contest’s running order, political climate and voting trends, and YouTube metrics, such as views and likes.
Last year she correctly predicted 15 out of 20 of the semi-finalists and successfully backed the winner, The Code by Switzerland’s Nemo, who beat favourites Baby Lasagna from Croatia. Adjustments to her model are already yielding improvements: she was right about 8/10 of the songs to qualify from Tuesday’s semi-final, including long shots SAn Marino and Portugal, against 7/10 last year.
She says she particularly enjoys trying to “beat the odds” in cases where her predictions don’t align with the bookies’.
“I liked The Code, I was happy it won,” says Polish-born Kasia. “Last year I placed a small bet on the winner and was pleasantly surprised — I might have a bit of fun with it again this year!”
Kasia says the Eurovision Song Contest is the “ideal testing ground” for her work.
“One of the hardest things about studying music popularity is choosing song datasets and defining success metrics,” she says. “Eurovision works really well as it provides a fixed list of songs each year and yields clear results so I can measure success. It’s also a fun and engaging project that allows me to share my research with a wider audience.”
As the music industry evolves with technology, Kasia hopes her insights into what makes a hit song might help her pursue a career in the business. “I interned at EMI records for a year and a half,” she says. “I think it would be really nice to get into the music industry.”
Kasia is studying for her PhD through the Centre for Digital Music, having completed an integrated master’s in Meng Electronics with Music and Audio Systems there in 2020. Her research is part of the University’s UKRI Centre for Doctoral Training in Artificial Intelligence and Music (AIM).
“Queen Mary is such a great place to study music and audio technology,” she says. “It's home to cutting-edge research in the field, and many of its researchers are actively involved in shaping the music industry through their work.” It has been wonderful to be able to focus on my interest in my studies and I hope I can continue to pursue it when I finish next year.”
Kasia appeared on BBC News ahead of the final to talk more about her model and predictions.