Analyzing Popularity Bias in Recommender Systems

Recommender systems often favor popular items, potentially reinforcing already popular items and reducing diversity. This thesis investigates popularity bias in recommendation algorithms by analyzing which types of items are most frequently recommended. Using several established datasets, we aim to implement baseline recommendation models and measure the distribution of recommended items across popularity levels. We also explore simple debiasing strategies—such as popularity-aware re-ranking or long-tail promotion—and compare their impact on accuracy and diversity.