Explorando Formas de Calibração e Redução do Viés de Popularidade em Sistemas de Recomendação

Recommender systems aim to enhance user engagement by suggesting new content based on their interests. Common algorithms like collaborative filtering recommend items similar to those users prefer, facilitating content discovery. However, recent research has identified issues such as unfairness, calibration errors, and popularity bias. While popularity bias promotes popular item consumption, it can lead to unfair recommendations that do not accurately reflect individual user interests. Current calibration methods focus on fairness but often overlook the amplifying effect of popularity bias. Our study addresses this gap by investigating calibration techniques and bias reduction strategies to deliver fairer recommendations aligned with user preferences, across various popularity levels. This research contributes insights into system calibration, fairness, user experience, and evaluation metrics.