Human-subjects evaluation of FairFeed onboarding and exposure correction

Investigate whether FairFeed’s onboarding interface recovers voters’ real interests with sufficient accuracy, whether transparent preference signals increase trust and perceived agency, and whether exposure correction remains inexpensive when real interests are narrower and more fragmented than the simulation’s stylized topic categories.

Background

FairFeed is evaluated through a simulation in which voters’ preferences are represented by broad topic categories rather than elicited from real participants. The paper therefore identifies a human-subjects deployment as the next step for assessing whether the proposed onboarding process can accurately translate citizens’ natural-language interests into preference signals used for ranking.

The evaluation is also needed to test behavioral and institutional assumptions that the simulation cannot establish, including whether transparent preference declarations improve users’ trust and sense of agency, and whether boosting under-exposed proposals remains compatible with relevance when actual preferences are more specific and fragmented than the model assumes.