Overview | People | Collaborators | Sponsors | Publications | Tools

The objective is to understand, identify and mitigate persuasion attempts by malign actors. In particular, we are concerned with the nature and process of persuasion, and the interplay between identity, affect and moral reasoning at both the cognitive and the social network level. The multi-disciplinary team (CMU, Buffalo, USMA, USC) blends the humanities, with the social and computational sciences to engage in a transdisciplinary program of research employing social media field studies, experimentation, machine learning and computer simulation to generate new findings, methods, tools and theory. This program is anchored by examining persuasion attempts in four scenarios defined by state, time frame and key event using data from multiple social media platforms including Twitter, Gab, YouTube, 4Chan and Reddit. These data are empirically analyzed to understand how presentation of moral values, affect, and identities in social media posts (verbally or visually) impacts the persuasiveness of the message and its ability to shape community and narrative. It is also used to test new tools for detecting identity and moral reasoning, to train and test new machine learning techniques for assessing moral values and persuasiveness, and to instantiate and validate an agent-based simulation model. This agent based model, Peitho, is used to understand the interplay between identity, affect and moral reasoning and to assess alternative courses of action for preventing or countering persuasion attempts by malign actors, and increasing individual and community resilience.

A data analysis pipeline is constructed and employed that spans from data gathering to counterfactual reasoning using simulation. Social media data are augmented using extant bot, troll, cyborg, user-social network position, frame and meme detection tools. Moral foundations theory, identity theory and affect control theory tools provide the basis to develop and employ novel methods for extracting feature vectors from the meta-data, images, and verbal content in social media posts. Dynamic statistical and network analytics are used to determine the relation of collected features to the messages persuasiveness. Virtual experiments are run to generate empirically testable hypotheses and counter-factual reasoning for course of action assessment.

There are five key contributions of this project. First, it integrates social psychology factors such as affect and moral foundations into a co-evolutionary perspective on social & knowledge networks. Second, it blends deep cognitive modeling with detailed social network modeling to create a social-cognitive framework for theory assessment and course of action analysis. Third, it develops new tools to analyze complex multi-platform, verbal and visual data to assess persuasiveness. Fourth it lays the ground work, methodologically and theoretically, for assessing memetic warfare. Fifth, it increases the utility, robustness and accuracy of the BEND framework by augmenting it with operationalized moral reasoning and identity components. This latter contribution can also be thought of as creating a scalable, operationalized technology for assessing the impact of moral reasoning and identity on persuasion as moderated by affect. Hence, this research lays the ground work for new techniques, tactics and procedures for: 1) rapidly identifying & reasoning about adversarial influence campaigns; 2) increasing community resilience to social-cyber-attacks; and 3) enabling authorities to more effectively improve social trust and protect their reputations and relationships. The result is a capability improvement that can be used in many areas from intelligence to public affairs. Approved for public release.