CAREER: Interacting Dynamic Bayesian Models for Social Behavior and Reasoning

Duke University, Durham NC

Investigators

Abstract

This project develops machine learning methods to analyze online data and illuminate social behavior. People spend a great deal of their lives socializing, or interacting with other people. Social interactions are inherently part of most of our activities, and, therefore, understanding social interactions is a fundamental part of understanding human behavior. As the amount of time people spend online rapidly grows, social interactions are increasingly occurring online. This project uses Dynamic Bayesian models to analyze time series data of social behavior that inherently involves interactions over time. The project provides tools for understanding social behavior and better methods for data analysis. The project trains graduate students and helps high school students learn more about data analysis. The project principal investigator also maintains a strong commitment to actively involving in the organization of Women in Machine Learning Workshop to ensuring the advancement of women in STEM. This research focuses on understanding the underlying structure of the interactions themselves. The research team elucidates this underlying structure through hierarchically modeling the interactions between multiple dynamic processes. More specifically, the project develops: (1) interacting dynamic Bayesian methods that can be used to model social interactions that jointly capture the temporal dynamics and the linguistic content of interactions between individuals, and discover latent attributes of individuals, such as power and influence, or roles such as bullies and victims; and (2) methods for analyzing epidemiological social network data.