Chatbot in the E-Service of Mental Health Using the Reprogramming of the GPT-2 Model

Abstract

Large language models LLMs, have revolutionized the field of natural language processing NLP by exhibiting human-like text generation capabilities. This paper explores the development of an advanced chatbot for mental health support using the GPT-2 language model. The paper aims to overcome the limitations of pre-trained models in domain-specific applications by implementing fine-tuning strategies and memory augmentation techniques. The methodology involves reprogramming the GPT-2 model using a dataset of mental health conversations with a focus on Alzheimer's disease. The system incorporates an array-based storage method for maintaining context across interactions, enabling more coherent and personalized responses. Evaluation metrics include BLEU scores, cosine similarity, and training loss to assess the quality and relevance of generated responses. Results demonstrate the model's ability to generate contextually appropriate and informative text on complex medical and social topics related to mental health. The paper highlights the potential of combining large language models with specialized memory approaches to enhance e-service of mental health interventions. Future work suggestions include expanding the training dataset, implementing the proposed system in other and different e-services fields. This paper contributes to the ongoing efforts to improve the accessibility and quality of mental health support through AI-driven conversational agents.