With vaccines and vaccine hesitancy all over the news, Assistant Professor Deena Abul-Fottouh’s co-authored paper, Examining algorithmic biases in YouTube’s recommendations of vaccine videos, makes for timely reading. The paper, which was published in August of 2020, concluded that YouTube’s policy of demonetizing “harmful content”, along with other changes to its recommender algorithm may have reduced the visibility of anti-vaccine videos.
Compared to an earlier 2016 paper by Prof Abul-Fottouh’s co-authors, Melodie Yunju Song and Anatoliy Gruzd, the number of recommendations for pro-vaccine videos has also significantly increased. However, the authors expressed concerns that users watching anti-vaccine videos are less likely to receive recommendations for pro-vaccine videos due to a “homophily effect” observed in the recommendation network.
The authors also suggest that public health agencies ought to collaborate with social media platforms to audit AI-driven recommendations so in order to address inherent biases built into recommendation models.
They are also working on another paper looking at COVID-specific misinformation related to vaccines.
