Reducing Bias in a Misinformation Classification Task with Value-Adaptive Instruction.

Instructional technology that supports the development of media literacy skills has garnered increased attention in the wake of recent misinformation campaigns. While critical, this work often ignores the role of myside bias in the acceptance and propagation of misinformation. Here we present results from an alternative approach that uses natural language processing to model the dynamic relationship between the user and the content they are consuming. This model powers a debiasing intervention in the context of a "fake news detection" task. Information about the user- and content-values was used to predict when the user may be prone to myside bias. The intervention resulted in significantly better performance on the misinformation classification task. These results support the development of content-general and embedded debiasing systems that could encourage informal learning and bias reduction in real-world contexts.