Key points are not available for this paper at this time.
Recommender systems play a crucial role in suggesting relevant content to users based on their past activities. They employ a process known as "collaborative filtering" to efficiently navigate extensive content repositories. However, concerns have been raised regarding the potential bias and homogeneity in recommendations, resulting in filter-bubbles and echo-chambers. Detecting and mitigating these biases is crucial for ensuring fair and diverse automated decision-making systems. This study investigates the impact of YouTube's recommendation algorithm on three distinct narratives across multiple dimensions. Our objective is to identify potential biases and gain insights into its decision-making behavior. We applied a multi-method approach to evaluate emotional content, moral foundations, lexical similarity, and social network analysis across 5 depths of YouTube recommendations. The results of our analysis showed diversity in emotions, significant drift in topics, and a push toward non-related, but highly influential videos across multiple recommendation depths. The findings from this study contribute to the understanding of bias in recommender systems. These insights inform the development of strategies to mitigate biases and improve the user experience. Policymakers and platform developers can utilize this knowledge to establish effective guidelines and policies for their recommender systems, enhancing decision-making processes.
Çakmak et al. (Mon,) studied this question.