On the heels of a study that examined political polarization on social media feeds, a fresh batch of research is shedding light on gender-based disparities. A report published in Cornell University‘s arXiv database shows that recommendation algorithms treat male and female accounts differently, especially in the realm of political content.
To run their experiment, the researchers created 160 virtual accounts. 80 of those accounts were seeded with “male-coded” categories like cars, sports, and gaming, while the other 80 received a female-coded base in categories like how-to and style. Each account also received a baseline seed of political content regardless of gender coding.
Then, the researchers turned on the recommendation pipeline. In total, the accounts were exposed to more than 500,000 algorithmic recs, and that sample revealed some modest — but significant — gender differences.
Subscribe for daily Tubefilter Top Stories
Since all accounts were seeded with political content, they were all recommended political content, but 56% of that content was served to female-coded accounts. The topics of those recs also varied based on gender; male-coded accounts were more likely to see content based around issues like crime and military defense, while female-coded accounts saw more posts related to international affairs and cultural concerns.
“Within-group similarity was consistently higher than between-group similarity, indicating that the recommendation system allocated different political content to different gender-coded profiles,” the researchers wrote. They also indicated that “male-coded users were steered toward a narrower set of confrontational domestic-order issues, whereas female-coded users were exposed to a broader informational environment populated by more multidimensional, moderate, and establishment-oriented macro and public-policy issues.”
Translation: The spread of political issues served to female-coded accounts was more varied, whereas male-coded accounts had more siloed experiences that honed in on certain issues. Gender differences on social media don’t just relate to the content of algorithmic recommendations, but to the diversity of those recs as well.
As with the study on political polarization, the gender study demonstrates the subtle forces that invisibly shape social accounts. It helps to be aware of these algorithmic differences — even if individual users may lack the agency to adjust their feeds.
