| With the advent of the era of big data,various information and various search engines emerge in an endless stream on the Internet.People can quickly use computers or smartphones to search for things they want to learn and understand anytime,anywhere.However,in addition to effective information in Internet search engines,It is often accompanied by a lot of redundant information,which can seriously damage the user experience,especially in the medical field.In addition,in most cases,the knowledge of related diseases will be searched on the Internet first,and self.diagnosis will be carried out.However,inaccurate and unprofessional information on the Internet will often cause great trouble to users and even delay the patient's most important medical treatment.It is unrealistic for users to effectively screen out correct and applicable medical knowledge from thousands of information on the Internet at the best time for treatment.In response to the above problems,this paper conducts research on entity recognition and intent analysis methods in the medical field and builds a medical intelligent question answering system based on the above research to help users retrieve medical knowledge efficiently and accurately.The main content of this paper has three aspects:(1)Named entity recognition task is one of the key research areas of natural language processing,but most of the previously named entity recognition tasks are based on English,and there is relatively little research on named entity recognition in the Chinese medical field,such as traditional RNN model,LSTM model,etc.Neural networks often perform poorly in the Chinese medical field due to their gradient disappearance,gradient explosion,and other problems.On this basis,this paper proposes an ALBert+Bi LSTM+CRF model,which can not only better adapt to medical texts that are generally long and difficult to learn context,but also solves the problems of large volume,many parameters and long training time of the Bert model,and compared with each model for the Chinese medical named entity recognition task.The comprehensive F1 value of the model on the medical data set reaches 91.40%,and it is in the leading position in the comparison experiments with various models.(2)The intent analysis task can be regarded as a multi-classification task,but there are very few datasets for intent analysis in the medical field.In this paper,the existing data is manually screened and labeled,and an intent analysis suitable for medical question answering systems is constructed.The dataset solves the problem of the shortage of datasets in this field and proposes an ALBert+Text CNN model,which makes up for the deficiency that the Text CNN model is insensitive to the context information and can only extract local features,and targets some drugs,diseases,etc.in medical texts.For the problem of long words,the size of the convolution kernel of Text CNN is expanded to 5 types.After comparative experiments,the accuracy of the model proposed in this paper reaches 90.24%,which is 1.52% higher than the basic Text CNN model.(3)Based on the above two studies,this paper constructs an intelligent question answering system based on question parsing.Firstly,the data crawling of the medical and health website is carried out,and a knowledge graph containing 44,000 entities and 294,000 entity relationships is constructed as the database of the question answering system;Secondly,a question parser is designed as the core module of the question answering system by integrating the above trained named entity recognition model and intention analysis model;Finally,using the flask development framework,a user-friendly front-end visual interface is designed to complete the construction of the intelligent question answering system. |
PDF Full Text Request