Modeling Heterogeneous User Behavior in Interactive Systems by Graphical Model and Collaborative Learning Framework

Abstract

In recent years, the rapid technological innovations of smart personal technologies have given rise to the growth of smart apps that can interact with users and implement personalized incentives to coordinate and change user behaviors in various realms such as e-commerce, patient-centered health system, and individual level transportation demand management (TDM) systems. Understanding user behaviors is crucial for further intervention strategy development and user experience optimization, hence the key to the success of the emerging applications. However, the existing statistical models encounter challenges when facing the unique characteristics of the systems, e.g., the user-system interactions make the apps more than data collection tools, but they also interfere with the user and change the user’s behavior; the users are heterogeneous in their preferences but data of a single user is limited and fragmented; the massive user base and its complicated structure will affect personalized learning and recommending. This dissertation develops novel models to address the aforementioned challenges based on collaborative learning framework, graphical models, and deep matrix factorization.