User Modeling for Adaptive Virtual Reality Experiences : Personalization from Behavioral and Physiological Time Series

Abstract: Research in human-computer interaction (HCI) has focused on designing technological systems that serve a beneficial purpose, offer intuitive interfaces, and adapt to a person's expectations, goals, and abilities. Nearly all digital services available in our daily lives have personalization capabilities, mainly due to the ubiquity of mobile devices and the progress that has been made in machine learning (ML) algorithms. Web, desktop, and smartphone applications inherently gather metrics from the system and users' activity to improve the attractiveness of their products and services. Meanwhile, the hardware, input interfaces, and algorithms currently under development guide the designs of upcoming interactive systems that may become pervasive in society, such as immersive virtual reality (VR) or physiological wearable sensing systems. These technological advancements have led to multiple questions regarding the personalization capabilities of modern visualization mediums and fine-grained body measurements. How does immersive VR enable new pathways for understanding the context in which a user interacts with a system? Can the user's behavioral and physiological data improve the accuracy of ML models estimating human factors? What are the challenges and risks of designing personalized systems that transcend current setups with a 2D-based display, touchscreen, keyboard, and mouse? This thesis provides insights into how human behavior and body responses can be incorporated into immersive VR applications to enable personalized adaptations in 3D virtual environments. The papers contribute frameworks and algorithms that harness multimodal time-series data and state-of-the-art ML classifiers in user-centered VR applications. The multimodal data include motion trajectories and body measurements from the user's brain and heart, which are used to capture responses elicited by virtual experiences. The ML algorithms exploit the temporality of large datasets to perform automatic data analysis and provide interpretable explanations about signals that correlate with the user's skill level or emotional states. Ultimately, this thesis provides an outlook on how the combination of recent hardware and algorithms may unlock unprecedented opportunities to create 3D experiences tailored to each user and to help them attain specific goals with VR-based systems, framed using the overarching topic of context-aware systems and discussing the ethical risks related to personalization based on behavioral and physiological time-series data in immersive VR experiences.