AutoPersuade: A framework for generating, evaluating and, revising arguments

AutoPersuade: A framework for generating, evaluating and, revising arguments

Princeton Docket # 24-4141-1

AutoPersuade is an innovative framework for generating, evaluating, and refining persuasive arguments. This technology offers a comprehensive solution for organizations and individuals seeking to create more effective messaging across various domains, including marketing, politics, and public health campaigns. Its ability to identify persuasive elements and predict the effectiveness of new arguments makes it a valuable tool for any organization looking to optimize its communication strategies. By providing actionable insights into what makes messages compelling, AutoPersuade enables users to craft more persuasive content and achieve better outcomes in their communication efforts.

The framework operates in three key steps: First, it curates a large collection of arguments and gathers human evaluations of their persuasiveness. Second, it employs a novel topic model called SUN (SUpervised semi-Non-negative) to extract the features of arguments that make them more or less persuasive. Finally, it uses the model output to evaluate the causal effects of argument features and forecast the persuasiveness of new arguments. This allows users to iteratively improve their messaging strategies based on data-driven insights

 

Applications

  • Optimizing marketing and advertising campaigns
  • Refining political messaging and public health communications
  • Analyzing and countering the spread of misinformation

 

Advantages

• Data-driven insights into what makes messages persuasive

• Ability to evaluate and predict effectiveness of new arguments

• Iterative improvement based on audience responses

 

 

Stage of development

The AutoPersuade framework has been developed and successfully applied to two diverse scenarios: promoting veganism and analyzing the spread of election conspiracy theories.

 

Inventors

Till Raphael Saenger is currently a Ph.D. student in the Operations Research and Financial Engineering Department. His research interests include statistical methodology, casual inference with text and applications of large language models.

Musashi Hinck Ph.D. Musashi Hinck completed his postdoc at Princeton University and currently works at Intel Labs as an AI research scientist.

Justin Grimmer Ph.D. is the Morris M. Doyle Centennial Professor in Public Policy in Stanford University's Department of Political Science and a Senior Fellow at the Hoover Institution. His research focuses on American political institutions, elections and voting behavior, and developing new machine learning and data science methods for studying politics.

Brandon M Stewart Ph.D. is an Associate Professor in the Department of Sociology at Princeton University. His research focuses on computational social science, particularly developing new statistical methods for analyzing large collections of text, with applications in political science and sociology.

 

Intellectual Property & Development status

Patent protection is pending.

Princeton is currently seeking commercial partners for the further development and commercialization of this opportunity.

 

Contact

Prabhpreet Gill

Princeton University Office of Technology Licensing • (609)258-3653 • psgill@princeton.edu

Patent Information: