A novel domain knowledge augmented large language model based medical conversation system for sustainable smart city development

The medical conversation system plays a significant role in smart healthcare services, contributing to the development of sustainable smart cities. While pre-trained large language models have exhibited impressive performance in general domain conversations, the effectiveness in medical applications remains limited due to the scarcity and inferior quality of domain-specific corpora. We introduce a domain knowledge augmented medical conversation system in Chinese, constructed on an advanced large language model with a reduced carbon footprint to support sustainable smart city development. The overall performance is enhanced through three key innovations, including retrieval augmentation from domain knowledge, dynamic masking training strategy, and prompt augmentation with question intention training data. To the best of our knowledge, this research represents initial efforts to integrate knowledge augmentation into a large language model-based medical conversation system in Chinese while actively controlling its environmental impact. Comparative experiments against state-of-the-art baseline models using comprehensive Chinese medical domain benchmark datasets demonstrate the efficiency of the proposed medical conversation system. The experimental results show remarkable improvements, with approximately 2% to 4% enhancement across evaluation metrics, underscoring the potential to facilitate smart healthcare services for sustainable smart cities. Future directions will concentrate on extending the Chinese domain-specific corpora, developing a multilingual medical conversation system, and designing efficient algorithms for refining the medical conversation system, which aims to reduce the carbon footprint while enhancing performance and generalization.