The filter bubble generated by artificial intelligence algorithms and the network dynamics of collective polarization on YouTube: the case of South Korea

Abstract

This study examined the role of the 'filter bubble,' an algorithm-mediated YouTube video suggestion system, in political polarization and the presence of echo chamber patterns in public engagement. We examined the mechanism by which automated YouTube recommendations augment selective exposure to ideologically similar content and the network-based dynamics of collective polarization. We collected lists of videos recommended by conservative and progressive news media and the accompanying replies using YouTube application programming interfaces (APIs) embedded in YouTube Data Tools and Webometric Analyst 2.0. The study examined similarities in content of related videos and conversation networks between commenters, as well as the relationships between the videos, channels, and commenters. Almost half of the videos had content similar to the original videos offered by both conservative and progressive news media. The filter bubbles in progressive media were not as strong as those in conservative media. More than 30% of the videos recommended by progressive media were from legacy media, while this figure was 25% for conservative sources. This study has significant implications for reverse engineering studies because it collected data through APIs, which are neutral and can access YouTube servers without logging into individual IDs.