Construction And Implementation Of Chinese Electronic Medical Record Knowledge Graph Based On BiLSTM

In recent years,related technologies in the medical field have developed rapidly,and medical health issues have become one of the most concerned issues.On the one hand,many websites of medical health spring up like mushrooms,resulting in more and more ways for people to seek medical treatment.With the rapid development of electronic medical health data,the way hospitals record patient information has changed from traditional paper medical records to computerized electronic medical records.On the other hand,based on its unique text characteristics and professionalism,electronic medical records are very difficult to construct training corpus manually,and there is no uniform labeling specification.Therefore,the model method of entity recognition and relationship extraction in the traditional field is difficult to apply in the electronic medical record,which poses a huge challenge to the natural language processing task in the medical field.To overcome these challenges,the construction and implementation of knowledge graph provides a suitable solution for the storage and management of medical knowledge.The paper designs a construction and implementation of knowledge graph based on deep learning.It performs named entity recognition and Relationship extraction on the text of medical electronic medical records,and then uses graph database to store medical knowledge and construct knowledge graph.The main contents of the paper are as follows:(1)In the method of medical named entity recognition,this paper designs the BiLSTM-CRF model for extracting features.The small sample annotation dataset is used to train the entity recognition model to extract the linguistic features and structural features of the electronic medical record.then continuously augment theannotation data set,and iteratively optimize the model repeatedly.The BiLSTM model solves the gradient disappearance problem of the traditional RNN model,and handles the long-term dependence of the RNN model through the control of“forgetting gate”,“input gate” and “output gate”.At the same time,the basic limitations of the maximum entropy Markov model based on the directed graph model and other Markov models are overcomed by the CRF model,which solves the problem of long-term dependence and label offset.(2)Aiming at the feature selection problem of relationship extraction between entities,this paper designs the BiLSTM-Attention model,and adds the BiLSTM layer to the Attention layer for entity relationship classification.The Attention layer first generates the weight vector of the sentence sequence,and then the lexical-level features into sentence-level features by multiplying the input vector by the weight vector of the sentence sequence,which reduces the information loss and information redundancy in the feature vector extraction process.(3)The extracted entities and entity relationships are represented by diagram model and stored in a graph database.The graph database uses an unstructured way to store data with complex associations and large depth of association,enabling efficient relational queries.This paper aims to design the construction and implementation of knowledge graph for unstructured medical texts through the above method,and enhance the semantic understanding ability in the process of constructing a medical knowledge graph through deep learning methods.The medical name entity recognition,relationship extraction and visualization of knowledge graph are elaborated in detail.We hope that these results can be further applied to a wider range of medical tasks to promote the research and development of knowledge in natural language processing.