Auditing YouTube's recommendation system for ideologically congenial, extreme, and problematic recommendations

Area of Science:

• Social Media Algorithms
• Algorithmic Bias
• Online Radicalization

Background:

• Social media algorithms face criticism for promoting ideologically aligned and extreme content.
• Evidence regarding filter bubbles and radicalization pathways remains inconclusive.

Purpose of the Study:

• To systematically audit YouTube's recommendation algorithm at scale.
• To investigate algorithmic recommendations for ideological congruence and extremity.
• To determine if recommendations become more extreme deeper in the user's watch history.

Main Methods:

• Utilized 100,000 sock puppet accounts for systematic algorithmic auditing.
• Analyzed recommendation patterns for ideological alignment and extremity.
• Tracked recommendations across increasing depths of the user watch trail.

Main Results:

• YouTube's algorithm recommends ideologically congenial content to partisan users.
• Congenial recommendations increase deeper in the watch trail, particularly for right-leaning users.
• While ideological extremity did not significantly increase, the proportion of problematic channel recommendations grew, especially for very-right users.

Conclusions:

• YouTube's algorithm favors ideologically congruent content, with some cross-cutting exposure possible.
• Problematic content, though low in proportion, is increasingly recommended deeper in the trail for right-leaning users.
• Algorithmic recommendations may contribute to exposure to fringe content for specific user groups.