Binge-Watching Your Way to Polarization? Measuring and Addressing Political Echo Chamber Effects on YouTube

Abstract

This project centers on the investigation of political polarization on social media, specifically the effects of automated content recommendation systems. The new research in this project is composed of three parts. In the first, we design and implement an experiment to simulate user activity on YouTube and collect data about the suggestions generated by YouTube’s proprietary recommendation algorithm. In the second part, we train a convolutional neural network to classify the videos that were generated in the experiment according to political/ideological orientation, and we analyze the results for evidence of the recommendation algorithms causing ideological drift or radicalization over time. In the third part, we design and build a Chrome extension to allow users to visualize their own YouTube watching history and the political implications of it, with the aim of allowing users to stay informed and avoid unwanted polarization. The results of the data analysis support the conclusion that conservative content dominates liberal content on YouTube, demonstrates that YouTube recommendations are more likely than not to align with the political leaning of the current video, and shows how the ideologically content of recommendations varies significantly with the topic of videos being watched. In the discussion, we move from the specifics of the YouTube project to broader implications for the intersection of social media and politics.