Using a prediction tool, Katarzyna Adamska, a third year PhD student, correctly predicted that Switzerland’s song entry, ‘The Code’, would win this year’s competition
The annual Eurovision Song Contest took place this May, with thirty-seven countries competing, and one winner chosen by televoting and votes from national juries. Katarzyna used machine learning to predict the outcomes of this year’s semi-finals and the grand final, identifying potential top performers and the overall winner – Switzerland.
Katarzyna studies within the AI & Music Centre for Doctoral Training and her research topic, ‘Predicting Hit Songs: A Multimodal and Data-Driven Approach,’ focuses on understanding the factors that contribute to a song's success, particularly in the UK Official Music Charts. She investigates various characteristics of songs from audio and musical perspectives to lyrics features, and measure song appeal through YouTube views and the frequency of use in user-generated content on social media. Thus, the Eurovision Song Contest offers a unique opportunity to apply her research in an exciting and highly visible setting.
Commenting on her prediction tool, Katarzyna said: “Multiple unpredictable factors influence Eurovision results. Political events throughout the year can sway countries to support or disfavour other nations, while the quality of live performances can boost the chances of underdog entries. Additionally, some countries have a history of voting for their neighbours or strategically supporting other nations. The competition also features a diverse range of genres and styles—from pop and rock to traditional regional sounds — which vary significantly from one participating country to another, year to year. This can make it challenging to discern a clear pattern of success.
The prediction tool I developed integrates several factors to accommodate some of the unpredictable elements of Eurovision results, for example, strategic voting behaviours among countries. In addition to previous voting patterns and the running order, the tool considers audio features, lyrics repetitiveness, and daily YouTube views to reflect both the intrinsic quality of the songs and their public appeal. Using this data, I employed four machine learning regression models to predict the rankings in the semi-finals and the grand final. I also experimented with various combinations of these data features to determine the most effective predictors for competition success. When interpreting results in scenarios like the Eurovision Song Contest, it's crucial to select a machine learning method that consistently delivers the best and most stable outcomes across all events, including both semi-finals and the grand final.”
To find out more about Katarzyna and her research, read her PhD profile