CRII: III: Generative Models for Robust Real-Time Analysis of Complex Dynamic Networks

University Of Toledo, Toledo OH

Investigators

Abstract

Many complex systems in the computer, information, biological, and social sciences can be represented as networks with nodes denoting objects and edges denoting relationships between the objects. Such complex network structures often change continuously over time through the observation of events at irregular times, such as networks of social interactions between people via messages, networks of transactions between organizations, and networks of face-to-face interactions between people. This project formulates a range of models of varying complexity for continuously evolving networks to enable robust real-time analysis of these networks in a variety of application settings. Such dynamic network models could be used in many scientific disciplines and in public health applications, including modeling the spread of airborne viruses between people. The project trains new graduate and undergraduate students, including female students from the University of Toledo's ACM-W chapter, in practical data science research involving a variety of data types and sources. The project also results in the development of an open-source Python software package, DyNetworkX, for analyzing dynamic networks along with educational materials on dynamic networks through a series of lectures and hands-on tutorials using the DyNetworkX package. This project aims to create a range of probabilistic generative models for continuous-time event-based networks that are flexible enough to account for the types of complex structures seen in real network data, including node popularity, community structure, reciprocity, and transitivity. The project also seeks to develop efficient incremental inference algorithms and discrete-time approximations that allow for real-time analysis of extremely large social networks that are rapidly changing over time, such as those seen in online social network data. The proposed range of models allows an analyst to trade off flexibility and scalability depending on the needs of a particular application. Two main applications are targeted: prediction of the spread of infectious disease over networks of physical proximity and real-time summarization and prediction of online social network activity. Deliverable assets of the project include new probabilistic models and inference algorithms, the DyNetworkX open-source software package, and educational materials on dynamic networks. These are intended to benefit researchers and educators in the computer and information sciences as well as researchers in other fields such as the social and economic sciences, software engineers, and hobbyists who work with dynamic network data. 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.