Research on hospital information management system based on large language model (LLM) algorithm model

Abstract

In recent years, the application of artificial intelligence, particularly generative AI (AIGC), has gained significant traction across various industries. This study focuses on employing a Large Language Model (LLM) algorithm to construct a generative dialogue model specifically tailored for traditional Chinese medicine, utilizing the extensive database of the Chinese Medical Code as the primary data source. The methodology consists of three key phases: (1) Corpus Collection and Preprocessing: This involves web crawling, data cleaning, and segmenting ancient texts to prepare a comprehensive dialogue dataset; (2) Model Training: A structured approach is applied through continuous pretraining using the Ziya-LLaMA-13B-v1 base, followed by supervised fine-tuning and DPO optimization; and (3) Evaluation: The model’s performance is assessed through automatic metrics such as BLEU (average score of 0.465) and ROUGE (average score of 0.392), alongside manual evaluations indicating high response accuracy (scores between 8 and 9). The results suggest that the model demonstrates robust generative capabilities in traditional Chinese medicine, outperforming existing models in answering knowledge questions. This study highlights the transformative potential of generative AI in healthcare, paving the way for enhanced hospital management and patient engagement tools.