Entity Recognition And Prediction Of TCM Knowledge Graph Based On Regular Expression And Neural Network

Clinical symptoms,as an important basis for traditional Chinese medicine differentiation and medication,contain rich patient characteristic information.At present,the symptoms in medical records are mainly described in natural language and traditional Chinese medicine language,which makes it difficult for computers to obtain patient symptom information in medical records.In response to the issue of effective utilization of symptoms in traditional Chinese medicine medical records,the thesis constructs a symptom data model and knowledge graph.Using the established fusion model and classification model,the symptoms in medical records are entity recognized and predicted for classification.The resulting symptom information is effectively classified and stored for visual display.The main work of the thesis is as follows:1.In response to the problem of symptom unit description and classification storage,the thesis analyzes and sorts out many symptoms in traditional Chinese medicine medical records,establishes a hierarchical symptom basic data model,divides symptom distribution,symptom location and symptom units into three layers and determine their subordinate relationships.Merge important elements of symptoms,discard factors that have less impact on symptoms,improve symptom units,and construct a basic knowledge graph of symptoms.2.Build a dataset for symptoms named entity recognition tasks in medical records.In addition to symptom custom entity annotation,add grammar rule annotation information.The RE-LSTM-CRF model is established and Bert is used as the pre training model.Based on the short-term memory network and conditional random field model,the regular expression discrimination unit and feedback unit are integrated into building the fusion model.Set up comparative experiments based on different types of conditions to verify the effectiveness of the fusion model under standards such as accuracy,accuracy,recall,F1 value,training time and model size.3.Predict and classify symptom descriptions and establish an Att TextCNN RF model.After processing the symptom named entity description and the additional filled description,the convolutional neural network with attention mechanism is used to extract the symptom description features and according to the different categories contained in the symptom part layer in the symptom basic knowledge map,the prediction classification of symptom entity description is completed in combination with the random forest model to complete the symptom basic knowledge map,And conduct comparative experiments with different classification models under the same conditions to verify the effectiveness of classification.4.Revise some symptom unit data in the symptom knowledge graph,improve the knowledge graph,and visualize symptom entity recognition and knowledge graph to facilitate the understanding and use of symptom information in downstream tasks of traditional Chinese medicine.This thesis provides an effective tool for obtaining symptom entities and classifying symptoms in the development of traditional Chinese medicine informatization by verifying the effectiveness of the established fusion model and classification model.The improved symptom data model and symptom knowledge graph provides basic data support for downstream tasks such as syndrome differentiation,diagnosis and treatment,and medication in the field of traditional Chinese medicine in the future.