Proceedings of the 4th International Conference on Public Management and Intelligent Society, PMIS 2024, 15–17 March 2024, Changsha, China

Research Article

Building a Medical Q&A System Based on Deep Learning and Knowledge Graphs

Download75 downloads
  • @INPROCEEDINGS{10.4108/eai.15-3-2024.2346400,
        author={Zhi  Wang},
        title={Building a Medical Q\&A System Based on Deep Learning and Knowledge Graphs},
        proceedings={Proceedings of the 4th International Conference on Public Management and Intelligent Society, PMIS 2024, 15--17 March 2024, Changsha, China},
        publisher={EAI},
        proceedings_a={PMIS},
        year={2024},
        month={6},
        keywords={knowledge graph; langchain; text2vec-chinese; chatglm-6b; qlora; localized medical q\&a},
        doi={10.4108/eai.15-3-2024.2346400}
    }
    
  • Zhi Wang
    Year: 2024
    Building a Medical Q&A System Based on Deep Learning and Knowledge Graphs
    PMIS
    EAI
    DOI: 10.4108/eai.15-3-2024.2346400
Zhi Wang1,*
  • 1: Hubei University of Technology
*Contact email: 1840782209@qq.com

Abstract

The aim of this paper is to build a localized question and answer system for medical big language models. Deep learning is used to equip the big language model with the ability to intelligently recognize and generate answers. In this paper, the completeness of knowledge storage of knowledge graph is used to provide the model with a more comprehensive knowledge of the medical specialized domain, which in turn is transformed into a dataset that can be used to train the model. In order to make the model more accurate, langchain is introduced as a new model prompt word design method, so that the new model inputs have identification and collect contextual information according to the established input index to expand the knowledge of the inputs, and then merge with the prompt to generate the inputs needed by the new model format, and the model makes use of the text2vec to deal with the Chinese participle in advance in the process of processing, so as to achieve a slight decoupling and through the comparison, the model is able to generate the answers in a more comprehensive way. The model uses text2vec to process Chinese word separation in advance during processing, achieving a slight decoupling and, by comparison, text2vec-english has a better effect on Chinese word separation.