# Auditing health-related recommendations in social media: A Case Study of Abortion on YouTube

**Authors:** Mohammed Lahsaini, Mohamed Lechiakh, Alexandre Maurer  
**arXiv:** [2404.07896](https://arxiv.org/abs/2404.07896)  
**Date:** 2024-04-11  
**Summary page:** https://www.summarizepaper.com/en/paper/2404.07896v1/

## Key points

- The study focused on recommendation systems (RS) employed by platforms like YouTube and their impact on information consumption, specifically regarding controversial topics such as abortion.
- A novel approach was introduced to scrutinize how YouTube recommends abortion-related content to individuals from diverse backgrounds.
- Analysis involved over 11,000 unique videos in YouTube recommendations for six simulated profiles seeking abortion-related content.
- Graph theory metrics such as In-degree, Weighted in-degree, Eigen centrality, PageRank, Katz centrality, and Authority were used to identify key videos influencing the network.
- The study highlighted the importance of using graph centrality measures for bias analysis and evaluating social influence within recommendation systems.
- YouTube tends to prioritize pro-abortion content over anti-abortion content in its recommendations while curbing misinformation spread on this topic.
- The research findings align with YouTube's efforts to combat misinformation surrounding abortion but could benefit from further validation through parallel studies conducted prior to the announcement.

## Summary

In this study, we delved into the realm of  (RS) employed by  like YouTube, with a particular focus on their impact on information consumption in the , specifically regarding controversial topics such as abortion. Recognizing the significance of  in uncovering potential bias and misinformation perpetuated by these algorithms, we introduced a novel  to scrutinize how YouTube recommends abortion-related content to individuals from diverse backgrounds. Our investigation involved analyzing over 11,000 unique videos that appeared in YouTube recommendations for six simulated profiles seeking abortion-related content. By constructing recommendation graphs for each profile and utilizing graph theory to assess centrality metrics such as In-degree, Weighted in-degree, Eigen centrality, PageRank, Katz centrality, and Authority, we were able to identify key videos that wielded significant influence within the network. This comprehensive approach not only enhanced the reliability of our rankings but also provided a nuanced understanding of the importance of individual videos based on various criteria. The utilization of graph centrality measures to pinpoint influential nodes within the network represents a commendable strategy for bias analysis and evaluating social influence. By identifying top-ranked videos that shape user interactions and information dissemination, we shed light on content that may either reinforce or challenge existing biases. Moreover, our optimized auditing process circumvented labeling bottlenecks typically associated with large video datasets, resulting in increased efficiency. Through our research findings, we observed that YouTube tends to prioritize pro-abortion content over anti-abortion content in its recommendations while effectively curbing the spread of misinformation on this topic. These insights align with YouTube's efforts to combat misinformation surrounding abortion announced in July 2022. However, further validation through parallel studies conducted prior to this announcement could enhance the credibility of these findings. Overall, our study underscores the importance of  in mitigating bias and misinformation within recommendation systems like YouTube's RS. By employing innovative approaches like  and leveraging graph theory metrics for analysis, we contribute valuable insights into understanding how these algorithms shape information consumption and influence user perspectives on contentious issues like abortion within digital spaces.