Designing Engaging Conversational Interactions for Health & Behavior Change

Abstract

The recent popularity of chat and voice-based conversational interactions fueled by advances in natural language processing (NLP) has opened up opportunities for re-imagining user interactions in health & behavior change as conversational experiences. Prior work has indicated that a well-designed conversational approach can be more engaging, motivating, natural, personal, and understandable. It can also mimic the properties of some of the most successful human-led interventions, such as coaching and motivational interviewing. However, designing conversational interactions poses numerous challenges. Efficiently creating conversational content that is diverse, relevant for the context, and sounds natural is challenging. Furthermore, balancing the still limited AI capabilities with user expectations requires careful problem scoping and other design considerations. Finally, the mechanisms in which a successful conversational interaction can help improve user engagement are still not well explored. In this dissertation I propose 4 different conversational systems that address some of the fundamental health & behavior change challenges. In Chapter 3 to address the intrinsic challenge of user boredom and engagement loss with repeated interactions - I propose a conversational system with value-based conversation topic personalization and diversification. In Chapter 4 to address the challenge of engaging users in mindful self-learning from their behavioral data - I propose conversational systems supporting structured reflection on physical activity and on professional development at work. In Chapter 5 to support health data collection, especially to improve user comfort in sensitive topics and understandability among low-literacy populations - I propose a system for conversational survey administration. Finally in Chapter 6, to lower the effort involved in designing good quality conversational systems, I propose a tool for automated conversion of form-based surveys to a more engaging conversational format. My work identifies and provides evidence for several benefits of the use of conversational interactions in health & behavior change. Among others, I demonstrate the benefits of increased engagement in interaction, improved motivation for performing activities, accessibility benefits related to familiarity, ease of use, comfort with sharing, and an ability to guide the users in the behavior change process via dialogue. I also identify several important challenges: perceptions of artificiality, managing high expectations of contextual knowledge, and social intelligence, as well as lower efficiency that could negatively affect the experience for some user groups. I further investigate the concrete links between conversational design elements and these benefits and challenges. My thesis demonstrates various design processes and automation techniques that can lower the effort of designing conversational experiences. As technology progresses conversational interactions can offer valuable support complimenting the existing automated tracking and the efforts of human health coaches. My work offers an important contribution to our understanding of how conversational interactions can play such a beneficial role.