Publication Details

Auditing YouTube's Recommendation Algorithm for Misinformation Filter Bubbles

SRBA, I.; MÓRO, R.; TOMLEIN, M.; PECHER, B.; ŠIMKO, J.; ŠTEFANCOVÁ, E.; KOMPAN, M.; HRČKOVÁ, A.; PODROUŽEK, J.; GAVORNÍK, A.; BIELIKOVÁ, M. Auditing YouTube's Recommendation Algorithm for Misinformation Filter Bubbles. ACM transactions on recommender systems, 2023, vol. 1, no. 1, p. 1-33. ISSN: 2770-6699.
Czech title
Auditovanie algoritmu odporúčania na YouTube pre dezinformačné filtračné bubliny
Type
journal article
Language
English
Authors
SRBA, I.
MÓRO, R.
TOMLEIN, M.
Pecher Branislav, Ing. (DCGM)
Šimko Jakub, doc. Ing., PhD. (DCGM)
ŠTEFANCOVÁ, E.
Kompan Michal, doc. Ing., PhD. (DCGM)
HRČKOVÁ, A.
PODROUŽEK, J.
GAVORNÍK, A.
Bieliková Mária, prof. Ing., Ph.D. (DCGM)
URL
Keywords

Audit, recommender systems, filter bubble, misinformation, personalization, automatic labeling, ethics, YouTube

Abstract

In this article, we present results of an auditing study performed over YouTube aimed at investigating how fast a user can get into a misinformation filter bubble, but also what it takes to burst the bubble, i.e., revert the bubble enclosure. We employ a sock puppet audit methodology, in which pre-programmed agents (acting as YouTube users) delve into misinformation filter bubbles by watching misinformation-promoting content. Then they try to burst the bubbles and reach more balanced recommendations by watching misinformation debunking content. We record search results, home page results, and recommendations for the watched videos. Overall, we recorded 17,405 unique videos, out of which we manually annotated 2,914 for the presence of misinformation. The labeled data was used to train a machine learning model classifying videos into three classes (promoting, debunking, neutral) with the accuracy of 0.82. We use the trained model to classify the remaining videos that would not be feasible to annotate manually. Using both the manually and automatically annotated data, we observe the misinformation bubble dynamics for a range of audited topics. Our key finding is that even though filter bubbles do not appear in some situations, when they do, it is possible to burst them by watching misinformation-debunking content (albeit it manifests differently from topic to topic). We also observe a sudden decrease of misinformation filter bubble effect when misinformation-debunking videos are watched after misinformation-promoting videos, suggesting a strong contextuality of recommendations. Finally, when comparing our results with a previous similar study, we do not observe significant improvements in the overall quantity of recommended misinformation content.

Published
2023
Pages
1–33
Journal
ACM transactions on recommender systems, vol. 1, no. 1, ISSN 2770-6699
Book
ACM Transactions on Recommender Systems
DOI
10.1145/3568392
BibTeX 
@article{BUT180410,
  author="SRBA, I. and MÓRO, R. and TOMLEIN, M. and PECHER, B. and ŠIMKO, J. and ŠTEFANCOVÁ, E. and KOMPAN, M. and HRČKOVÁ, A. and PODROUŽEK, J. and GAVORNÍK, A. and BIELIKOVÁ, M.",
  title="Auditing YouTube's Recommendation Algorithm for Misinformation Filter Bubbles",
  journal="ACM transactions on recommender systems",
  year="2023",
  volume="1",
  number="1",
  pages="1--33",
  doi="10.1145/3568392",
  issn="2770-6699",
  url="https://dl.acm.org/doi/10.1145/3568392"
}
Back to top