Research And Implementation Of Medical Registration System And Question Answering Matching Model

With the development of Internet and the rapid growth of network data,there are many medical communities,more and more people begin to make information acquisition and consultation in the medical QA community.Medical communities use online doctors to provide medical services,but also use accumulated data to provide information retrieval services,but mainly based on search engine keyword matching to return a bunch of related questions and answer documents,unable to deeply understand the semantic information of user questions.QA system has become a research hotspot because it can return a certain answer rather than a bunch of documents that still need to be filtered by users.The question answering system based on deep learning regards the question answering process as a process of semantic matching between question sentences and candidate answers.At present,there are two main research directions: using multiple network combinations to enhance the ability of network feature extraction;using attention mechanism to identify the intention of questions,highlighting the interactive information between questions and answers.However,when using multiple combinations of deep neural networks to extract the deep features of question and answer sentences,the features of the last layer of the network are often used as sentence features,instead of the features of the middle layers;when using attention mechanism to capture the information related to the answers,the final question features are also used to weight the answer features,without considering each phase in the middle Interactive information of questions and answers between the same layer.In view of the fact that patients are limited by their own medical knowledge when they go to the hospital to register and determine the Department,sometimes they don't know the Department they need to hang;similarly,when users consult through the medical community,they also need to select doctors according to the Department category(such as internal medicine,facial features department,obstetrics and Gynecology),but some users can't determine which department their disease belongs to.As far as I know,there is no relevant research so far.In view of the above problems,the main research work of this paper is as follows:(1)In this paper,deep learning and natural language processing technology are applied to solve the problem of medical registration for the first time,and the problem of choosing department for patients' medical registration is modeled as the problem of question classification.Based on the data accumulated by the medica l community,this paper uses Bi-LSTM and attention mechanism to train a classifier to classify patients' problems.In addition,LSI text similarity technology is used to find the most similar problem in labeled data sets,and the result of classifier is verified according to the label of the most similar problem.The registration system can facilitate users to confirm the registration department and find the doctors they need in the online medical community.(2)In this paper,we propose a multi-level fusion model of question and answer matching,which uses the depth neural network of multi-level feature fusion and the attention mechanism of multi-level to solve the question and answer matching problem.The features extracted from each layer of Bi LSTM + CNN + CNN are spliced and fused as the final question and answer semantic features for question and answer matching and answer scoring.Attention mechanism is added to each layer of question and answer feature extraction to highlight the answer and question related features.Compared with the features of the last layer of the deep neural network,the model uses the features of each layer synthetically,and the semantic information of the final features is more sufficient;the attention mechanism is used to capture the interactive information of question and answer in time between each same layer.(3)In view of the complex processing of Chinese word segmentation and removal of stop words when using word vectors,in addition,although Chinese word segmentation tools are relatively mature now,due to the existence of professional terms in specific medical fields,some professional word segmentation recognition is not allowed to directly affect the downstream model performance,while building a professional field dictionary can solve this problem,but the task is large and is poor in transplanting to other fields.The above problems can be avoided by using char vectors which needn't word segmentation.Therefore,in this paper,in the process of question answering semantic matching,in addition to using word vector for model training,we also use word vector for experiments to study the effectiveness and feasibility of word vector.(4)Using the data of medical community,registration model and question answering matching model,a medical question answering system is designed to solve the problem of user registration and question answering needs.Compared with the medical community information retrieval service,its question answer module is based on a bunch of questions and corresponding answer documents related to keyword matching return and user questions.The question answer system designed in this paper uses the multi-layer fusion feature interactive question and answer matching model to score the questions and candidate answers after deep semantic matching,and returns a certain answer.