Understanding Social Dynamics Through Coevolving Latent Space Networks With Attractors

Sponsor: National Science Foundation

Award Number: 2120115

PI: Eric Kolaczyk

Co-Is/Co-PIs: Dylan Walker, Konstantinos Spiliopoulos, Dino Christenson

This research project will develop a general class of coevolving network models. In social systems, interactions frequently influence individual behavior and beliefs which can, in turn, impact interactions. Specific variants of this type of coevolutionary phenomenon include opinion dynamics, voter behavior, observational learning, herding or flocking, and polarization. Network-based models are natural for representing such phenomena, and relevant work can be found in both the mathematical and statistical literatures (among others). However, coevolving network models are substantially less well-developed than models for networks of other types (e.g., static networks) and they are more complex to analyze and understand. This project will develop a model class that integrates central elements of the mathematical and statistical coevolving network modeling literatures. The models will be used to examine polarization in two online social network data sets, Twitter for Congress and Reddit. The project will involve a collaboration between statistical, mathematical, political, and computational social scientists. Graduate students will receive cross-disciplinary training in these areas. Publicly available software will be developed.

This research project will develop a general new class of coevolving latent space network with attractors (CLSNA) models for social systems. The development of the CLSNA model class will result in a new type of causal modeling framework, explicitly combining dynamical systems modeling from mathematics with hierarchical modeling and inference from statistics. The former will allow the investigators to incorporate mathematically precise notions of social dynamics, like attraction and repulsion. The latter will permit computationally tractable and theoretically supported methods for statistical inference. In this project, the investigators will: (i) develop the modeling and statistical inference methodology for CLSNA models, with an emphasis on flocking and polarization; (ii) study the resulting behaviors allowed by this class, through a combination of both numerical and mathematical techniques; and (iii) assess empirically in online social media data the nature and extent of specific coevolutionary behaviors using these models. The models to be developed will be general and quite broadly applicable. The investigators, however, plan to focus initial applications on the context of affective polarization using online social network data sets.

For more information, click here.