Empowering users to control news algorithms reduces filter bubbles but not for all

A new study published in the proceedings of the AAAI Conference on Web and Social Media investigates whether giving users direct control over their news recommendation algorithms can help break filter bubbles. The researchers built a political news recommendation system augmented with an interface that transparently shows the political stance and topic interests the system has inferred from a user's reading behavior. Users could then adjust sliders to receive more articles on a particular topic or from a particular political leaning. This design contrasts with typical opaque recommendation systems where users are unaware of how their preferences are being shaped.

In a controlled user study comparing the transparent system to a traditional interface, the researchers found that the enhanced interface significantly increased users' awareness of being in a filter bubble. However, the effects on actual news consumption were mixed. While many users moved the system from extreme liberal or conservative positions toward the center, this came at the cost of reduced political diversity in the articles shown. Moreover, a subset of users chose to move the system to even more extreme positions, amplifying their own filter bubble. The findings indicate that empowering users with control does not automatically reduce polarization; rather, it depends heavily on individual preferences and goals. The study underscores the need for careful design of user-centric recommendation systems that balance transparency, control, and diversity.