Creating an Automated IdeologicalTransformer Using Moral Reframing

Abstract

This work proposes a novel ideological transformer tool, called ShiftView, based on the moral reframing concept, by applying natural language processing tools and techniques. Liberals and conservatives have different morals based on the Moral Foundations Theory. For example, liberals are more likely to base their judgments on the moral value of harm while conservatives may base their judgments on the moral value of purity. Studies have shown that while people can convince others of the same political ideology, they use the same moral rhetoric to convince people with different ideologies, and ultimately fail to do so. Mapping moral rhetoric to match a different ideology is a powerful persuasive tool called moral reframing. This thesis seeks to create ShiftView, an automated ideological transformer using moral reframing. This project has four parts: generating original datasets from the records of the United States Congress and Twitter, designing a flexible moral classifier using topic modeling, creating a text generator using Recursive Neural Networks and transfer learning, and developing an ideological transformation framework. During the process, several moral reframing methods were prototyped and evaluated. Initial results are reviewed by Amazon Turk Workers. This research is the first step in the multi-year project of automating ideological transform and moral reframing.