Decision Flow Networks for Effective Classification in Service Systems

Northwestern University, Evanston IL

Investigators

Abstract

This award advances the security of online communications by improving the process of verifying information posted to social media outlets in near real-time. Whether through cyber-attacks or inaccuracies, misinformation can have devastating effects on civil society and preventing its spread is considered by Congress to be one of the major security challenges facing the Nation today. Because today's social media networks are so highly connected, information can spread through social media at great speed. As information is made public and shared online, the challenge facing online media outlets is to balance the accuracy and speed of news verification in order to avoid epidemics of misinformation. This project will develop new methods to direct internal (proprietary scoring algorithms) and external (third-party fact-checking) resources in order to detect and disrupt misinformation quickly and accurately. The research opens new directions in time-sensitive classification problems and has wide application to time sensitive decision-trees involving binary decisions. This project will provide opportunities for graduate students to learn about cutting-edge security challenges through interactions with online media firms. Employing a new paradigm that integrates a decision-making framework with queueing-type performance measures, this project studies decision flow networks. This paradigm extends standard queueing networks by allowing a sequence of decisions to impact knowledge flow and vice versa. Decisions depend heavily on the random processing times and accuracy of agents (e.g., algorithms or fact-checkers) who provide service to the network. Focusing on decision flow networks with binary decision (e.g., jobs are classified into one of two categories), this award develops novel operations research models that capture the interdependence between the agent's knowledge and the flow of information in the network. These models integrate decision theory and queueing theory, and the analysis require a mix of robust optimization and stochastic dynamic programming techniques. This novel approach makes a rich contribution to the existing literature and extends the boundaries of each of these fields. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.