Multimodal Health Chatbot Interface for Doctors and Patients in Dementia Care
Abstract
Dementia progression is tracked through trends in longitudinal signals that clinicians must reconstruct from fragmented sources before each consultation. Home health-monitoring devices continuously capture much of this data, yet clinicians cannot query it at the point of care. This thesis investigates where conversational access to a patient’s longitudinal record adds value for clinicians and where it adds effort over a visual overview. Using a design-science approach, a clinician-facing interface was co-designed with two memory clinic specialists. The prototype pairs a dashboard with a Rasa-based conversational drill-down that provides descriptive responses, and the co-design yielded a reusable catalog of 56 clinician-validated queries across 15 content intents. It was evaluated quantitatively on its intent pipeline and summatively with three specialists outside the design process. The pipeline proved feasible (content-intent accuracy 0.883; named-metric extraction F1 0.965), but the harder problem was correctness, not reliability: a query can be confidently mapped to the wrong intent, observed once here yet intrinsic to the layered design. The central finding was dashboard primacy: all three specialists treated the dashboard as primary and the chat as secondary, but disagreed on the chat’s residual role beneath it, which we read as evidence that this role is better configured than fixed. Interpretability was the main challenge and drove a preference for inline chart markers over conversation; trust hinged on rendering accuracy and provenance; and the out-of-loop specialists contested several design decisions. The thesis thus contributes scoped, evidence-based design knowledge on when conversational access to longitudinal health data is warranted and when a visual overview is preferable, drawing a boundary rather than delivering a validated product from a prototype evaluated with three participants on synthetic data.
| Attribute | Value |
|---|---|
| Title (de) | Multimodaler Health Chatbot Schnittstelle für Ärzte und Patienten in der Demenzpflege |
| Title (en) | Multimodal Health Chatbot Interface for Doctors and Patients in Dementia Care |
| Project | |
| Type | Master's Thesis |
| Status | finished |
| Student | Jonathan Parth |
| Advisor | Joshua Oehms |
| Supervisor | Prof. Dr. Florian Matthes |
| Start Date | 08.01.2026 |
| Sebis Contributor Agreement signed on | 02.12.2025 |
| Checklist filled | Yes |
| Submission date | 08.07.2026 |