FLLMM: A Federated Large-small Language Model Collaboration Based Music Therapy for Mental Disease

Music therapy is important in the treatment of mental disease, but most existing methods rely on the synergy of domain experts. In this paper, we propose a federated large-small model collaboration based music therapy for mental disease (FLLMM). First, a pre-trained large language model (LLM) is introduced for fine-tuning to realize automated psychological suggestions and reduce the dependence on experts; second, a mechanism of large and small language model synergistic training is adopted, with the large language model using a priori knowledge to guide the small language model in the early stage, and the small language model supplementing the generalized knowledge of the large model in the later stage; lastly, an audio classifier (AC) and an audio knowledge base are designed to realize automatic recognition and mapping of audio categories and provide corresponding music therapy suggestions. The proposed method achieves the best results on the publicly available dataset mental_health_chatbot_dataset, and multiple ablation experiments are designed to verify the rationality of the model structure.