The eDEM‑Connect project, funded under grant 16SV8335 and carried out from 1 February 2020 to 31 July 2023, aimed to create a chatbot‑based communication and service platform for caregivers of people with dementia (PwD) dealing with behavioural problems such as agitation. A central element of the platform is a domain ontology that formalises knowledge about agitation and the stability of the caregiver‑patient relationship, providing the semantic backbone for question‑answer generation and intelligent dialogue.

The ontology development followed a four‑phase workflow. First, a comprehensive domain analysis was performed in close collaboration with the German Center for Neurodegenerative Diseases (DZNE) in Witten and Fraunhofer ISST. This phase involved requirement gathering, literature review, and interviews with experts to identify key concepts and relationships. Second, an initial ontology was conceptualised and implemented in the Web Ontology Language (OWL) using the Protégé editor. The initial version was validated through automated logical inference checks in Protégé and by domain experts from DZNE and the University of Witten/Herdecke, who assessed completeness and expressiveness. Third, the ontology was enriched by integrating existing knowledge bases and by applying natural language processing (NLP) techniques to extract additional concepts and relations from a corpus of publicly available forum posts. The corpus was scraped, anonymised, and annotated with high‑level concepts from the ontology. Annotation guidelines were published, and several annotators contributed to an iterative refinement process, resulting in a publicly available, reusable dataset. Fourth, the ontology was further validated by testing the chatbot’s ability to retrieve correct answers, in partnership with Fraunhofer ISST, which provided the dialogue system. The validation cycle led to iterative improvements and a more complete ontology.

The extraction pipeline demonstrated the feasibility of combining expert knowledge with data‑driven methods. Automated extraction techniques were applied to identify behavioural patterns, particularly agitation, within the textual data. Although the report does not provide quantitative performance metrics, the successful integration of extracted knowledge into the ontology and the subsequent improvement in chatbot response accuracy indicate that the approach is effective for this specialised domain. The ontology itself is designed to be reusable beyond the immediate project, enabling its application in proactive, situation‑adaptive assistive systems, smart‑home technologies, and wearable aids for dementia care.

Collaboration was a key driver of the project’s success. The University of Rostock led the technical work, coordinating the development of the initial ontology, its implementation, and the integration of new knowledge. The University of Greifswald served as the funding recipient and oversaw the overall project management. Partners from the University of Witten/Herdecke and DZNE Witten contributed domain expertise, while Fraunhofer ISST supplied the chatbot platform and facilitated validation. Additional collaborators included Gute Hoffnung and the University of Witten/Herdecke’s research group, all of whom participated in workshops and joint reviews to ensure the ontology met clinical and technical requirements. The project’s timeline was structured around four major milestones, each aligned with the phases of ontology development and validation, culminating in a final report that documents the ontology’s readiness for deployment in caregiver support systems.