Computer-aided applications based on artificial intelligence (AI) methods have become an indispensable part of everyday life. Their constant use by many users and providers leads to numerous facilitations. Examples can be found in the facilitation of information searches and in the support of decision-making processes.
The typical and everyday use of AI is for example, in music streaming services such as “Spotify” or “Last.fm”. The streaming services make use of so-called recommendation systems, which are used for searching, sorting, and filtering extensive music content. AI-generated music recommendations are based on training data corresponding to users’ past listening habits. This can lead to the unintended popularity bias effect, whereby music tracks by popular artists are overrepresented in the training data.
The Know Center is investigating the popularity bias effect in the field of music recommendation systems. Findings show that listeners of popular, mainstream music receive significantly more accurate recommendations. Listeners of unpopular, or alternative, music experience a disadvantage. The characteristics of the unfairly treated music listeners were analyzed using a variety of data science methods, such as clustering. It was discovered that openness to unfamiliar music has a positive effect on the accuracy of recommendations. These findings may be useful for developing unbiased music recommendation systems that meet the requirements of trustworthy AI.
Innovations in the field of recommendation systems and developments towards fairness will further help to increase trust in AI-based solutions. This will also meet the proposal by the European Union to make regulated and trustworthy AI a legal standard.