A Conversational Application for Insomnia Treatment: Leveraging the ChatGLM-LoRA Model for Cognitive Behavioral Therapy

The aim of this study was to develop a mobile application for psychotherapy with insomnia patients using the ChatGLM-LoRA model, fine-tuned by Low-Rank Adaptation, and validated in a clinical trial.The dataset used to train the model was a collection of 764 dialogues related to sleep disorders. The corpus was randomly divided into three subsets: training, validation, and test sets. The hyperparameters used in this study to train the model were 450 epochs, betas ranging from 0.9 to 0.95, weight decay rate 5e-4, maximum learning rate 1e-5, and AdamW optimizer. Based on the test results of the above hyperparameters, the four metrics of BLEU-4, ROUGE-1, ROUGE-2, and ROUGE-L of the model reached 0.0340, 0.0451, and 0.0163; 0.2773, 0.3075, and 0.1986; 0.0592, 0.0735, and 0.0261; 0.2112, 0.2336, and 0.1500 for the training, validation, and test sets.These results indicate the technical feasibility and potential clinical utility of using an advanced language model-based application for psychotherapeutic intervention in insomnia.