Given the reach of YouTube as a proliferator of contemporary news and ideas, it is important to understand how YouTube's machine learning recommendation algorithms reflect or reinforce pre-existing political bias. Previous research has examined user data and ideological homogeneity within social media groups, but the extent of YouTube's reflection of political bias remains relatively unexplored. Principally, our research aims to use natural language Processing techniques to provide a novel understanding of YouTube's reflection of political bias within its search and video recommendation algorithms. We created two experiments to understand each of the aforementioned systems. Experiment 1 examines the relationship between videos' ranking and political biases. We quantified such bias by applying an optimized BERT Natural Language Processing regression model to video transcripts. Experiment 2 examines the progression of bias when repeatedly clicking the “Up-Next” recommended video per each video cycle. We find that while the average bias of videos ranked highly in searches is slightly Democratic-leaning, YouTube surprisingly minimizes the magnitude of bias within their “Up-Next” recommendations. Ultimately, our results provide a nuanced understanding of YouTube's reflection of political bias and introduce an ethical discussion regarding the fairness of YouTube's algorithm.
Paper
Full text
Examining Political Bias within YouTube Search and Recommendation Algorithms
Semantic Scholar · Political Science · 2021
Abstract
Given the reach of YouTube as a proliferator of contemporary news and ideas, it is important to understand how YouTube's machine learning recommendation algorithms reflect or reinforce pre-existing political bias. Previous research has examined user data and ideological homogeneity within social media groups, but the extent of YouTube's reflection of political bias remains relatively unexplored. Principally, our research aims to use natural language Processing techniques to provide a novel understanding of YouTube's reflection of political bias within its search and video recommendation algorithms. We created two experiments to understand each of the aforementioned systems. Experiment 1 examines the relationship between videos' ranking and political biases. We quantified such bias by applying an optimized BERT Natural Language Processing regression model to video transcripts. Experiment 2 examines the progression of bias when repeatedly clicking the “Up-Next” recommended video per each video cycle. We find that while the average bias of videos ranked highly in searches is slightly Democratic-leaning, YouTube surprisingly minimizes the magnitude of bias within their “Up-Next” recommendations. Ultimately, our results provide a nuanced understanding of YouTube's reflection of political bias and introduce an ethical discussion regarding the fairness of YouTube's algorithm.