The Political Narrative of YouTube Recommendation System: 2018 Ontario election

Alt Tab

McKelvey, F., Frizzera, L. (2019) The Political Narrative of YouTube Recommendation System: 2018 Ontario election. Canadian Communication Association (CCA). Vancouver, Canada.

Visualization available here: http://labs.fluxo.art.br/rankflow/

Check our findings for the Brazilian Federal Elections: https://luciano.fluxo.art.br/algorithms-and-disinformation-the-role-of-youtube-in-the-brazilian-political-scene/

Social media is both an important tool for politicians to disseminate their platform and a way citizens learn about politics (Marland, 2016; Small, Jansen, Bastien, Giasson, & Koop, 2014). Algorithmic recommendation increasingly mediates this relationship in what is called a discoverability paradigm (Desjardins, 2016). Discoverability refers to the ways content discovery p latforms coordinate the relationships between citizens and politicians that make political content more or less engaging. While algorithmic recommendations can be helpful for citizens to find relevant content, algorithmic recommendations have been criticized as recommending extremist content or by not disclosing the partisan bias of certain information sources (P. Lewis, 2018; Tufekci, 2018).  

There is, however, a lack of data about political discoverability in Canada. This paper contributes to a first look through an important case of YouTube recommendations about party leaders during the 2018 Ontario election. YouTube is a top social media platform in Canada, used by 22% of Canadians for news according to the Reuters Institute Digital News Report. Most parties upload speeches, advertisements and exclusive content to YouTube. Finally,  YouTube’s recommendations drive 71% of viewing time on the platform (Covington, Adams, & Sargin, 2016; Solsman, 2018).

Our presentation makes three contributions:

  1. We prototype new methods to analyze YouTube by tracking the recommendations per party leader.
  2. We find that recommendations do not typically promote extremist content but recommendations more frequently promoting negative videos for Liberal leader Kathleen Wynne and positive videos for PC leader Doug Ford.
  3. We describe the next steps for political discoverability research. We recommend moving away from single source recommendations towards developing bot profiles to investigate the nature of the personalized recommendation.

We arrived at these conclusions by tracking recommendations over time from before the writ was drawn to election day (April 3 to June 2, 2018) using methods developed by the AlgoTransparency Project. We analyzed recommended videos by coding for tone towards party leader as well as developing visualization techniques to uncover the pathway through content and the changing popularity of recommended videos over time.

Our contributions matter for the study of politics in Canada because they qualify the influence of algorithmic recommendations on the accessibility of political content to information-seeking citizens. With much concern oppose the effects of filter bubbles or algorithmic radicalization, we instead see clear parallels between public opinion and video recommendations. However, we also discovered popular accounts seemingly outside mainstream public opinion, suggesting that alternative influence networks on YouTube might be an effective way of amplifying political messages or perspectives (R. Lewis, 2018).

References

Covington, P., Adams, J., & Sargin, E. (2016). Deep Neural Networks for YouTube Recommendations. Proceedings of the 10th ACM Conference on Recommender Systems. Retrieved from https://research.google.com/pubs/pub45530.html

Desjardins, D. (2016). Discoverability: Toward a Common Frame of Reference. Canada Media Fund. Retrieved from http://trends.cmf-fmc.ca/research-reports/discoverability-part-2-the-audience-journey

Lewis, P. (2018, February 2). “Fiction is outperforming reality”: how YouTube’s algorithm distorts truth. Guardian. Retrieved from http://www.theguardian.com/technology/2018/feb/02/how-youtubes-algorithm-distorts-truth

Lewis, R. (2018). Alternative Influence: Broadcasting the Reactionary Right on YouTube. Data & Society Research Institute. Retrieved from https://datasociety.net/output/alternative-influence/

Marland, A. (2016). Brand Command: Canadian Politics and Democracy in the Age of Message Control. Vancouver: UBC Press.

Small, T. A., Jansen, H., Bastien, F., Giasson, T., & Koop, R. (2014). Online Political Activity in Canada: The Hype and the Facts. Canadian Parliamentary Review, 37(4), 9–16.

Solsman, J. (2018, January 10). CES 2018: YouTube’s AI recommendations drive 70 percent of viewing [CNET]. Retrieved October 26, 2018, from https://www.cnet.com/news/youtube-ces-2018-neal-mohan/

Tufekci, Z. (2018, March 10). YouTube, the Great Radicalizer. New York Times. Retrieved from https://www.nytimes.com/2018/03/10/opinion/sunday/youtube-politics-radical.html

1 thought on “The Political Narrative of YouTube Recommendation System: 2018 Ontario election

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.

%d bloggers like this: