Research on Medical Dialogue Generation of External Knowledge

Abstract

Abstract Nowadays, the technology of medical dialogue generation has gradually attracted the attention of more researchers, and the demand for landing has gradually increased. Therefore, building a medical dialogue system that can automatically reply is conducive to improving the efficiency of clinical consultation and reducing the burden on doctors. This paper uses the method of fusing external knowledge to build a dialogue generation model, which greatly enhance the accuracy of the model and ameliorates the disadvantages of classical construction methods. Based on the large-scale pre-training model method, the doctor’s response is generated by two-stage training, and the knowledge related to the medical background is added to generate the response that best fits the current context. In this paper, experiments were performed on the medical dialogue dataset KaMed and COVID-19, and the experimental data showed that compared with the traditional human-computer dialogue generation Seq2Seq model, the Perplexity value of this method decreased 1.91, compared with the VHRED model, B@1 value increased 0.3, and the B@2 value increased0.34, D@2 increased 2.14, it can be proved that the medical dialogue model proposed in this paper can provide doctors with response responses more effectively and enhance the accuracy of responses.