Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

YouTube Recommendations Reinforce Negative Emotions: Auditing Algorithmic Bias with Emotionally-Agentic Sock Puppets

Hussam Habib, Rishab Nithyanand

TL;DR

The paper investigates whether YouTube's recommendation algorithm reinforces users' emotional preferences by auditing its behavior with emotionally labeled sock puppets. It reveals that the system amplifies negative emotions and that reinforcement strengthens over time and across contexts, with contextual recommendations sometimes more reinforcing than personalized ones. By comparing prevalence and prominence of emotionally aligned content, the study shows that YouTube can propagate emotional biases and create filter-bubble-like dynamics, raising concerns for user well-being and societal impact. The work highlights the need to balance personalization with content diversity and user agency, and it discusses methodological limitations and implications for platform design and policy.

Abstract

Personalized recommendation algorithms, like those on YouTube, significantly shape online content consumption. These systems aim to maximize engagement by learning users' preferences and aligning content accordingly but may unintentionally reinforce impulsive and emotional biases. Using a sock-puppet audit methodology, this study examines YouTube's capacity to recognize and reinforce emotional preferences. Simulated user accounts with assigned emotional preferences navigate the platform, selecting videos that align with their assigned preferences and recording subsequent recommendations. Our findings reveal reveal that YouTube amplifies negative emotions, such as anger and grievance, by increasing their prevalence and prominence in recommendations. This reinforcement intensifies over time and persists across contexts. Surprisingly, contextual recommendations often exceed personalized ones in reinforcing emotional alignment. These findings suggest the algorithm amplifies user biases, contributing to emotional filter bubbles and raising concerns about user well-being and societal impacts. The study emphasizes the need for balancing personalization with content diversity and user agency.

YouTube Recommendations Reinforce Negative Emotions: Auditing Algorithmic Bias with Emotionally-Agentic Sock Puppets

TL;DR

The paper investigates whether YouTube's recommendation algorithm reinforces users' emotional preferences by auditing its behavior with emotionally labeled sock puppets. It reveals that the system amplifies negative emotions and that reinforcement strengthens over time and across contexts, with contextual recommendations sometimes more reinforcing than personalized ones. By comparing prevalence and prominence of emotionally aligned content, the study shows that YouTube can propagate emotional biases and create filter-bubble-like dynamics, raising concerns for user well-being and societal impact. The work highlights the need to balance personalization with content diversity and user agency, and it discusses methodological limitations and implications for platform design and policy.

Abstract

Personalized recommendation algorithms, like those on YouTube, significantly shape online content consumption. These systems aim to maximize engagement by learning users' preferences and aligning content accordingly but may unintentionally reinforce impulsive and emotional biases. Using a sock-puppet audit methodology, this study examines YouTube's capacity to recognize and reinforce emotional preferences. Simulated user accounts with assigned emotional preferences navigate the platform, selecting videos that align with their assigned preferences and recording subsequent recommendations. Our findings reveal reveal that YouTube amplifies negative emotions, such as anger and grievance, by increasing their prevalence and prominence in recommendations. This reinforcement intensifies over time and persists across contexts. Surprisingly, contextual recommendations often exceed personalized ones in reinforcing emotional alignment. These findings suggest the algorithm amplifies user biases, contributing to emotional filter bubbles and raising concerns about user well-being and societal impacts. The study emphasizes the need for balancing personalization with content diversity and user agency.
Paper Structure (47 sections, 3 equations, 3 figures, 6 tables)

This paper contains 47 sections, 3 equations, 3 figures, 6 tables.

Table of Contents

  1. Introduction
  2. Background
  3. YouTube's Recommendation Algorithm
  4. How do recommendation algorithms optimize for engagement?
  5. Emotional preferences in consuming videos.
  6. Emotions can shape users' media consumption patterns.
  7. Related works
  8. Auditing YouTube’s recommendation algorithm.
  9. Emotional content, virality, and emotional contagion.
  10. Experiment Design
  11. Creating sock puppets.
  12. Assigning preference ($p$) to sock-puppets.
  13. Selecting seed videos ($s$)
  14. Revealing assigned preferences.
  15. Recording recommendations
  16. ...and 32 more sections

Figures (3)

  • Figure 1: Distribution of utility values derived from Up Next recommendations for preference-aligned (treatment) and control sock puppets, grouped by seed domain. Note that some utility values have been scaled independently better visual representation within the graph. For exact values refer to \ref{['tab:h1.1:results']}.
  • Figure 2: Distribution of utility values from personalized and contextual Up Next recommendations. We aggregate recommendations across all domains. Note that some utility values have been scaled independently better visual representation within the graph. For exact values refer to \ref{['tab:h2.1:results']}.
  • Figure 3: Mean utility within Up Next recommendations of predefined videos for preference-revealing (treatment) and random-selecting (control) sock puppets. The values are shown as the percentage difference from the first control-observed recommendations. All values at the 0th essentially show utility within contextual recommendations.