A carregar...
Publicação
FlexiDialogue: Integrating dialogue trees for mental health with large language models
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 813.23 KB | Adobe PDF |
Orientador(es)
Resumo(s)
The increasing prevalence of mental health issues among university students is exacerbated by limited access to support due to shortages of mental health professionals and the stigma associated with seeking help. Virtual mental health assistants can extend the reach of existing resources, but traditional systems reliant on scripted dialogues are constrained by inflexibility and limited adaptability to diverse user inputs. This paper introduces FlexiDialogue, a system that transforms rigid dialogue trees into instruction sets for large language models, facilitating dynamic, contextually appropriate, and multilingual interactions while maintaining the structure and quality of expert-validated dialogue flows. The system was evaluated in three phases: (1) determining how effectively large language models could map open-ended user responses to predefined dialogue tree options, allowing for more natural interaction without compromising control; (2) assessing the models’ ability to paraphrase scripted dialogues to improve conversational fluidity while remaining grounded in the original tree; and (3) conducting an expert review to assess overall performance. Results demonstrated that FlexiDialogue enhanced the flexibility and coherence of interactions, with expert evaluations confirming its potential for mental health support.
Descrição
Palavras-chave
Mental health virtual assistants Dialogue systems Large language models Natural language understanding Flexible dialogue trees Mental health support Multilingual interaction Conversational AI
Contexto Educativo
Citação
Editora
SCITEPRESS - Science and Technology Publications