The technologies enabling today’s social, economic, and communication networks are generating unprecedented amounts of behavioral data as users interact with devices and applications, e.g., fine-granular clickstream sequences from users viewing webpage contents.
The availability of this data presents opportunities for new approaches to network science through innovations in data science, optimization, machine learning, and related areas. In this talk, I will present two groups of network data science methodologies I have developed: data-driven efficiency optimizers (DEO) and behavior-based early detectors (BED). DEO involves modeling and optimizing the efficiency of networks, through joint maximization of the match
between network topologies and intended functionalities over possibly millions of variables corresponding to node properties and edge weights. BED is concerned with leveraging behavioral data to enable short timescale predictions of adverse events in networks, through pattern mining algorithms that identify signals within sequences of user actions and machine learning techniques that leverage these signals as features to minimize detection times. In describing BED and DEO, I will focus primarily on the modeling and optimization of Social
Learning Networks (SLN), which are types of networks that form between instructors, students, and modules of content. In doing so, I will include results from deploying DEO and BED to real-world SLN through my company Zoomi Inc.