Background Information Assisted SEQ2SEQ Automatic Question Answering System Of Medical Guide Station

As the modern medical system matures,more and more emerging technologies are used in medical systems.Among them,the automatic question and answer technology will bring great help to the doctor-patient communication.Generative question-and-answer systems based on deep learning generally use the Sequence to Sequence(Seq2Seq)framework to build question-and-answer models,but the model cannot make full use of the textual information in related scenarios,and some generated sentences are relatively simple.To make the responses generated by the model more abundant,this paper uses a new question-answering system that combines medical background information with the Seq2Seq question-answer model.The paper first studies the Chinese corpus preprocessing technology and the expression and training methods of Chinese word vectors.The word2vec training tool is used to train distributed word vectors.Compared with the sparse word vector representation method,the word vector dimension is reduced and the network calculation is reduced.Secondly,the Seq2Seq question answering model combined with the Attention mechanism is studied.In the model,a variety of recurrent neural networks are used as encoders and decoders.Compared with multiple sets of experiments,the network with the best model training effect is selected as the codec of the question answering model.Then,in order to improve the accuracy and diversity of responses generated by the question and answer model,a medical background information model is added to the Seq2Seq question and answer model;the Seq2Seq question and answer model is used as a pre-training model,and background information is added for retraining,so that the model not only related to the question and answer sentence,it also fully learned the characteristics of the background information,and the generated reply is more rich and realistic.Finally,because the traditional Q&A model evaluation index does not involve the semantic level of the sentence,which is different from the manual evaluation index,this paper uses a comprehensive evaluation index of relevance and similarity(RSEB).The evaluation index combines the similarity between the generated reply and the reference reply and the correlation between the generated reply and the question to comprehensively evaluate the model reply sentence.Through the Tensorflow deep learning algorithm framework,an automatic question and answer model is built to compare the response effects of the background information model and the traditional question answering model.The experimental results show that the semantics generated by the model proposed in this paper are richer and more complete,and the quality is higher than the responses generated by the traditional question answering model.Therefore,the automatic question answering model assisted by background information is more suitable for hospital guidance desks.And through the comparison of model test results,it is found that the RSEB index used in this paper has more advantages than the PPL index.