Investigating Popularity Bias Amplification in Recommender Systems Employed in the Entertainment Domain

Popular content gets recommended disproportionately more often: The research found that recommender systems systematically favor already-popular items, meaning songs, movies, and shows that are already widely watched get recommended even more frequently, while niche content gets overlooked.

This creates unfair experiences for users with diverse tastes: People who prefer less mainstream entertainment receive significantly less accurate recommendations compared to users who like popular content, meaning the system actively works against those seeking variety or specialized interests.

The problem spans multiple entertainment platforms: The bias amplification effect was consistent across music, movies, and anime domains, suggesting this is a systemic issue affecting how AI recommends content across the entertainment industry, not isolated to one platform or medium.

Current AI systems may violate fairness regulations: The paper connects popularity bias to compliance concerns under regulations like the European AI Act, which requires AI systems to address bias and fairness—suggesting many deployed recommender systems may not meet emerging legal standards for trustworthy AI.

Algorithm design choices matter for fairness: By studying the relationship between recommendation accuracy, how well-calibrated algorithms are, and popularity bias, the research highlights that developers have choices in how they build these systems that can either reduce or amplify unfair treatment of niche content and diverse users.