![Sample our Humanities journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/5939/TOC-Subject-Sample-2021-Humanities.jpg"})
ABSTRACT
Recommendation algorithms that customize information feeds for individuals have raised concerns about exacerbating inequalities in news exposure among citizens. In response to these concerns, we conducted an audit study on YouTube to analyze the algorithmic impact on curating news versus other content topics. We examined over 1.7 million YouTube video recommendations audited in 2019 and developed novel analysis approaches including network analysis and Markov chains. Results show that recommendation algorithms may potentially redirect users away from news content through two influence pathways: (1) the “topical filter bubbles,” wherein entertainment content has a higher probability of being recommended over news content in a self-reinforcing manner; and (2) “algorithmic redirection,” wherein the probability of entertainment videos being recommended after a news video is much higher than that for the opposite. Overall, YouTube recommendation algorithms have a higher probability of recommending entertainment videos than news. The findings imply essential biases in algorithmic recommendations on digital platforms beyond amplifying users’ preferences.
Acknowledgments
We thank Camille Roth for pending credit for us to reuse the data and share the original video IDs for validation and robustness checking. The original video IDs are respectfully kept in high confidence. We thank Biying Wu-Ouyang and Renyi He for assisting validation process. We are also grateful for the feedback and support from Michael Delli Carpini, Yphtach Lelkes, Yilang Peng, Subhayan Mukerjee, Emily Falk, Kevin Munger, Danaë Metaxa, Chenyan Jia, and colleagues from the Annenberg School for Communication at the University of Pennsylvania.
Disclosure Statement
No potential conflict of interest was reported by the author(s).
Supplementary material
Supplemental data for this article can be accessed on the publisher’s website at https://doi.org/10.1080/10584609.2024.2343769
Data Availability Statement
The data that support the findings of this study are available from the corresponding author, upon reasonable request.
Open scholarship
This article has earned the Center for Open Science badge for Open Data. The data are openly accessible at https://doi.org/10.1080/10584609.2024.2343769
Notes
1. We are aware of current debates around the blurring boundaries between news and entertainment in the hybrid media environment (e.g., Edgerly & Vraga, Citation2020; Williams & Delli Carpini, Citation2011). In this study, “news” is defined in accordance with a more conservative perspective: the political information that is vital for informed citizenship and representative democracy (e.g., Delli Carpini & Keeter, Citation1996).
2. The term “topics” (“topical”) in this paper refer to broader content categories, such as news, entertainment, music, movie, and games.
4. See https://indianexpress.com/article/explained/youtube-recommendations-explained-7587523/ and https://blog.hootsuite.com/how-the-youtube-algorithm-works/#A_brief_history_of_the_YouTube_algorithm
5. The unidentified data and analysis scripts can be accessed from the following link: https://osf.io/r85p9/?view_only=930b97a05ae5474fa7a5ba01419afaa3.
6. In an audit study of Google News, Haim et al. (Citation2018) found the personalized recommendation probabilities for news/politics, entertainment, and sports were 52%, 17%, and 33%, respectively. Compared to these results, YouTube shows higher probabilities of recommending the video with the same categories, which indicates stronger topical filter bubble effects.
7. Considering the low proportion of certain categories in the sample, which may result in a relatively low accuracy for the measure of transition probabilities, we group “pet,” “automobile,” “sports,” and “travel” into a single group, “others”
Additional information
Notes on contributors
Shengchun Huang
Shengchun Huang is a Ph.D. Candidate at Annenberg School for Communication, University of Pennsylvania. Her research interests include digital news consumption, algorithmic effects, and users’ perceptions of personalized media environments.
Tian Yang
Tian Yang is an Assistant Professor at the School of Journalism and Communication, Chinese University of Hong Kong. His research interest is at the intersection of digital media, political communication, and computational social science.