Designing text-based Chatbots for eating disorders : from an identity-based perspective

Abstract

Eating Disorders (EDs) are mental illnesses which result in various negative consequences. However, many individuals who suffer from EDs are not receiving professional treatment due to various reasons such as the lack of awareness, the feeling of shame, an attachment to the illness, financial constraints, and limited treatment capabilities. To address this situation, self-help programs have been developed using different media including manuals, CD-ROMs, websites, and mobile applications. Although existing studies in the field of EDs have indicated that patients’ multiple identities such as race, ethnicity, age, and gender interact with one another which impacts the risk, diagnosis, and treatment of EDs, existing self-help interventions for EDs have not emphasized the discrepancy among different user identities. To tackle this issue, this research adopts an identity-based perspective to study the complexities of user identities and analyze user needs associated with identities. In particular, the intersectionality-informed methodology is applied which examines how experiences are produced from multiple intersecting identities. Among diverse platforms/media, this research focuses on chatbots which are able to mimic human conversations to communicate with users and deliver interventions. Overall, this research aims to develop a framework for designing chatbots for people with EDs from an identity-based perspective. To achieve this objective, two major research questions are investigated. The first research question – aiming to understand user needs – is framed as: which identities impact people’s Eating Disorders and result in a variety of needs? The second research question – aiming to design for users – is framed as: how can a design framework for chatbots be developed to meet various user needs for people with Eating Disorders? This research project is comprised of three studies to respond to research inquiries. The first study is an unobtrusive online observational study on a question-and-answer platform. This study investigates which identities and how identities affect ED patients/potential patients. The results of the first study form a theoretical framework of identities which includes three types of possible identities (group identities, role identities, Eating Disorder identities), the relationship patterns among identities, and the relevance between different identities and user needs. The second study applies the theoretical framework from the first study and takes a further step, aiming to undertake a comprehensive analysis of individuals’ needs associated with their identities and create chatbot prototypes for user testing. This study consists of case studies which are structured as a four-step process, including data collection through qualitative methods, chatbot creation based on user identities, user testing, and analyzing user feedback. Findings of this study contribute to establishing a preliminary design framework for creating identity-based chatbots for EDs. This framework is built with four dimensions, including content, persona cues, conversational cues, and structure. Following a similar study design to the second study, the third study involves case studies to evaluate the design framework established in the second study in terms of user trust and chatbot usefulness. Chatbot prototypes for user testing are modified based on user feedback from the previous study. Factors that contribute to building a trustworthy chatbot are identified as a result of this study, involving chatbot-related factors, environment-related factors, and user-related factors.