Principles for Designing Context-Aware Applications for Physical Activity Promotion

Abstract

Mobile devices with embedded sensors have become commonplace, carried by billions of people worldwide. Their potential to influence positive health behaviors such as physical activity in people is just starting to be realized. Two critical ingredients, an accurate understanding of human behavior and use of that knowledge for building computational models, underpin all emerging behavior change applications. Early research prototypes suggest that such applications would facilitate people to make difficult decisions to manage their complex behaviors. However, the progress towards building real-world systems that support behavior change has been much slower than expected. The extreme diversity in real-world contextual conditions and user characteristics has prevented the conception of systems that scale and support end-users’ goals.

We believe that solutions to the many challenges of designing context-aware systems for behavior change exist in three areas: building behavior models amenable to computational reasoning, designing better tools to improve our understanding of human behavior, and developing new applications that scale existing ways of achieving behavior change. With physical activity as its focus, this thesis addresses some crucial challenges that can move the field forward.

Specifically, this thesis provides the notion of sweet spots, a phenomenological account of how people make and execute their physical activity plans. The key contribution of this concept is in its potential to improve the predictability of computational models supporting physical activity planning. To further improve our understanding of the dynamic nature of human behavior, we designed and built Heed, a low-cost, distributed and situated self-reporting device. Heed’s single-purpose and situated nature proved its use as the preferred device for self-reporting in many contexts. We finally present a crowdsourcing system that leverages expert knowledge to write personalized behavior change messages for large-scale context-aware applications.