Understanding Social Network-Transcendent Online-Offline Behavioral Dynamics- From Data to Models to Prediction
Abstract
The availability of data, coupled with advances in machine learning, has revolutionized engineering and computer science. An ecosystem of industry and academic collaboration in fields such as computer vision, reinforcement learning, deep neural networks, robotics, autonomous systems and AI have been an engine for economic growth over the last decade. Less apparent is the impact of data science and machine learning on the social sciences, especially on our ability to understand and predict social unrest and conflict, and to enable decision making to address the root causes of such events. However, a quiet revolution has also been taking place in the social sciences as well. Previously, social science theories were developed using limited amount of data, and thus it was difficult to validate or falsify such theories. Recent experiences have shown that it is important to incorporate data science and machine learning into the social sciences to better understand the role of online and offline behavior on events that impact society. For example, social scientists initial understanding of the 2010-12 Arab Spring uprisings, drawn from limited data, turned out to be substantially wrong when further data became available, thus demonstrating that data science can play a significant role in the social sciences. With this motivation, our goal in this proposal is to accelerate the convergence of data, network and social sciences by collecting, curating and interpreting online and offline behavioral data from multiple time and spatial scales, and to use network science and machine learning to design systems to enable social scientists to predict events that lead to unrest in society, politics and the government. From the social science perspective, the success of this project will provide advanced deep learning methods for correlating large-scale online-offline datasets at different scales and with different granularity and will result in new nonlinear, multi-scale, multi-network models for opinion- behavior dynamics. These computational and mathematical tools will offer unprecedented opportunities for social scientists to understand a broad range of online-offline interactions, e.g., characterizing the policy topic domains, types of grievance, and important dimensions of moralizing language associated with online-offline behaviors; and verifying four possible hypotheses to characterize online-offline relationships- the independence hypothesis, the facilitation hypothesis, the spillover hypothesis, and the reciprocity hypothesis. From the network science perspective, this project will lead to nonlinear, multi-scale, multinetwork models for opinion dynamics, which generalize existing single-scale, linear network models. The nonlinear feedback dynamics, and the analytical tractability of the model, make all the difference in being able to develop new systematic and principled means to predict rapid cascades of engagement over multi-scale, multi-network systems and to reveal the mechanisms that explain, and can be used to tune, sensitivity of these cascades to spatial and temporal location of internal or external cues. From the data science perspective, this project will generalize deep graph neural networks and recurrent neural networks in the context of multi-layered networks and tensors, will develop algorithms and tools that produce semantically meaningful, low-dimensional, and nonlinear representation of the multi-layered networks, and will provide real-time early warning algorithms for multi-scale, multi-layered networks by synthesizing statistical learning and big-data analytics.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 05, 2025
- Source ID
- FA95502410002
Entities
People
- Lei Ying
Organizations
- Air Force Office of Scientific Research
- Board of Regents of the University of Michigan
- United States Air Force