Research On The Generation Method Of Medical Question And Answer Answer Based On Automatic Abstract

The medical field has always been an important direction of informatization development.With the leap-forward development of medical informatization,the type and scale of medical data has shown a spurt of growth,and the data explosion has made medical care truly enter the era of big data.However,search engines return a series of related results,without considering the differences in the needs of people with different habits and interests.Compared with the traditional search engine,only a series of related documents can be fed back to the user.The question and answer system enable the user to input the question in natural language and give the user a simple and accurate answer instead of a series of related documents,which suggested that compared with the search engine,the question and answer system is more convenient and accurate.Thus,the question and answer system is an urgent need in the medical field.Through the analysis of the medical question and answer corpus,we can see that the category label of the medical question and answer corpus has this hierarchical feature,which brings challenges to the next answer generation.Therefore,the paper puts the problem classification and answer generation of medical question and answer corpus as the main research point of this thesis,and proposes the classification of medical problems based on hierarchical multi-label and the method of generating medical answers based on automatic summarization.details as follows:The medical problem is constructed in a tree structure,and there is an imbalance in the label distribution.In the leaf node,there are more samples under some labels but a few samples under other labels.Even under the high-level nodes,there is a phenomenon in which the label distribution is uneven,but the overall sample size is still large.This paper combines the advantages of deep learning and traditional machines learning to design a hierarchical classification model based on hybrid neural networks models.In the high-level node part,the problem text composed of the word vector matrix is used as input,and the convolutional neural network is combined with the deep belief network as the classifier of the high-level node to maintain the dependency between the high-level nodes.In this paper,the SVM is used as the classifier of the underlying node for the characteristics of the leaf nodes in the medical question and answer that there are more samples under some labels and fewer samples under other labels.Experiments show that the proposed method has a better performance on hierarchical multi-label classification than the simple CNN or SVM.Aiming at the complexity of the problem sentence structure and the existence of multiple semantics in the question of medical users,this paper proposes an automatic summarization based answer generation method,which calculates the topic probability vector of the problem text and calculates the similarity among problem texts.After obtaining the original answer set composed of the answerquestion pairs,which is similar to the user questions,the improved sequence-tosequence learning model based on the encoding-decoding structure,which is constructed by the Recurrent neural networks,is used to generate the answer sentences from the original answer set.Experiments show that the model in this paper has a certain degree of improvement in the accuracy of automatic question and answer generation.Finally,the medical automatic question answering system was designed and implemented.At the same time,the hierarchical multi-label based medical problem classification technology and the automatic abstract based medical question answer generation technology were integrated into the automatic question answering system.