Predicting Bias in the Evaluation of Unlabeled Political Arguments

AbstractWhile many solutions to the apparent civic online reasoning deficit have been put forth, few consider how reasoning is often moderated by the dynamic relationship between the user's values and the values latent in the online content they are consuming. The current experiment leverages Moral Foundations Theory and Distributed Dictionary Representations to develop a method for measuring the alignment between an individual's values and the values latent in text content. This new measure of alignment was predictive of bias in an argument evaluation task, such that higher alignment was associated with higher ratings of argument strength. Finally, we discuss how these results support the development of adaptive interventions that could provide real-time feedback when an individual may be most susceptible to bias.


Return to previous page