Improving the language model on the grounds of order of mental scales

Linguistic models are developed with the user's description or are used to predict the user's preferences and opinions. However, the existing methods of creating a person are highly dependent on the user's characteristics and / or previous judgments, but not on any reasoning underlying the user's judgments. We present PB & J (Psychology of Behavior and Judgment), a framework that improves LMs manually by including the possible reasons why a user makes a certain decision. Our reasons are modeled in a language to clearly communicate user behavior based on their experiences, personality traits, or beliefs. Our approach uses psychological scaffolds: A systematic framework such as the Big 5 personality traits or Primal world beliefs to help ground the generated ideas. Social opinion tests and social prediction tasks In addition, our PB & J PBs work in competition with those using human-authored methods, suggesting the power of artificial intelligence guided by existing ideas.

• † University of California
• ** Work Done while at Apple