Algorithmic audit reveals shifting news exposure from state to non-state media with user engagement, indicating asymmetric platform responsiveness to preferences.
Key Points
To investigate how recommendation algorithms allocate visibility between government-affiliated and non-government news sources across stages of user interaction and preference formation.
Conducted an agent-based testing (ABT) audit on Douyin to track news allocation dynamics across different user engagement stages.
Government-affiliated media receive greater visibility during early recommendation phases when user behavioral signals are sparse.
Non-government-affiliated media steadily gain visibility over time under sustained user interaction and engagement.
The recommendation platform exhibits asymmetric responsiveness depending on the specific type of user preference expressed.