Ελληνικό Μεσογειακό Πανεπιστήμιο
Conversational agents for empathetic interaction and sentiment analysis: a proof of concept study in palliative care for hematologic malignancies
Abstract
dc:descriptionConversational agents are increasingly explored as digital tools for supporting patient monitoring and communication in healthcare. Patients with hematologic malignancies often require long-term follow-up and frequently experience both physical symptoms and emotional distress. At the same time, clinical consultations are limited in time, making continuous monitoring between visits challenging. These conditions create an opportunity for conversational systems that can assist in collecting patient-reported information while maintaining an appropriate communication style. This thesis investigates the design and development of emotionally aware conversational agents for monitoring patients with hematologic malignancies. The research is based on the creation and analysis of a corpus of real Greek clinician–patient conversations collected during hospital visits. The dataset was manually transcribed and anonymized and used to study the thematic structure and emotional dynamics of clinical dialogue through topic modeling and sentiment analysis techniques. Building on these insights, the thesis explores the development of conversational agents in several stages. First, an intent-based conversational agent was designed to support structured data collection, including symptom reporting and family history information. Usability evaluation using the User Experience Questionnaire (UEQ) indicated that conversational interfaces can serve as acceptable and understandable tools for patient monitoring. Subsequently, more advanced conversational architectures were developed using large language models. A sentiment-aware conversational agent was implemented using a fine-tuned language model trained on clinical dialogue data. In this system, emotional signals detected in patient language influence dialogue flow, allowing the system to pause structured questioning and provide supportive responses when distress is detected. Expert-based evaluation using the CARE questionnaire indicated that the system was perceived as empathetic while maintaining a positive user experience. Finally, the thesis proposes an architecture that integrates large language model dialogue generation with retrieval-augmented generation (RAG) to ground responses on biomedical evidence and reduce hallucination risks. The system also supports longitudinal monitoring through clinician-facing dashboards that allow healthcare professionals to review patient-reported information over time. Overall, the findings suggest that conversational agents can support structured patient monitoring while maintaining emotional sensitivity in communication. The thesis contributes methodological insights and practical system designs for integrating conversational AI into clinical monitoring workflows, particularly in low-resource language environments such as Greek.
Degree
thesis:*- Grantor dc:publisher
- Ελληνικό Μεσογειακό Πανεπιστήμιο
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Chatzimina, Maria Evangelia
- Χατζημηνά, Μαρία-Ευαγγελία
Subjects
dc:subject × 33- Συνομιλιακοί πράκτορες
- Ενσυναισθητική αλληλεπίδραση
- Ανάλυση συναισθήματος
- Ανακουφιστική φροντίδα
- Αιματολογικές κακοήθειες
- Μεγάλα γλωσσικά μοντέλα (LLMs)
- Κλινικά συστήματα διαλόγου
- Επεξεργασία φυσικής γλώσσας στα ελληνικά
- Συναισθηματική υπολογιστική
- Παραγωγή επαυξημένη με ανάκτηση πληροφοριών
- Τεχνητή νοημοσύνη
- Conversational agents
- Empathy
- Sentiment analysis
- Palliative care
- Hematologic malignancies
- Large Language Models (LLMs)
- Clinical dialogue systems
- Greek language
- Retrieval Augmented Generation
- Affective computing
- Artificial intelligence
- Επιστήμη Ηλεκτρονικών Υπολογιστών και Πληροφορική
- Φυσικές Επιστήμες
- Διάδραση ανθρώπου-υπολογιστή
- Βιοϊατρική Μηχανική
- Επιστήμες Μηχανικού και Τεχνολογία
- Computer and Information Sciences
- Natural Sciences
- Human-Computer Interaction
- Medical Engineering
- Engineering and Technology
- Biomedical Engineering
Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/62449