Design And Implementation Of Application System Based On Medical Knowledge Graph

With the rapid development of artificial intelligence technology and the maturity of knowledge graph application technology in the vertical field,knowledge graph has broad application prospects in the medical field.Medical knowledge graph can help doctors make decisions,help the general public understand disease knowledge,and help patients understand the disease.The value of knowledge graph in the medical field is not only a medical knowledge base,but more importantly,it provides functions such as medical intelligent search.The basis of auxiliary diagnosis and treatment.With the improvement of medical informatization and the development of big data,the medical field has accumulated a large amount of underutilized data.Constructing a medical knowledge graph can fully tap the potential value of the data.This paper studies the process and method of constructing medical knowledge graph,uses deep learning,graph neural network and other technologies for medical entity extraction;builds knowledge query based on medical knowledge graph,automatic question answering of knowledge graph,and online medical entity extraction platform.The work of this paper mainly has the following points:(1)The label attention mechanism is studied,and the attention mechanism is applied to sequence labels to learn the embedded representation of label information.For the traditional LSTM entity extraction model,only the input sequence can be feature extracted without considering the label features.In this paper,a layer-by-layer improved label attention mechanism-based network LAN(Hierarchically-Refined Label Attention Network)is used.The model uses LSTM Feature extraction is performed on the input sequence,and the attention mechanism is used to learn the label features.The attention mechanism can capture the dependency relationship between label contexts.Using the CCKS2018 medical entity recognition competition data set in the medical NER task,it has been experimentally proved that the recognition accuracy of the BiLSTM-LAN model is higher than the BiLSTM-CRF model,but the results are slightly worse than the GGNN model.(2)Research on graph neural network technology and text sequence representation method in graph network.In view of the shortcomings of uncertain word boundaries and complex composition of Chinese named entity recognition,this paper studies the application of graph neural network in entity recognition model,and uses dictionary-based gated graph neural network(GGNN)to construct Chinese medical entity recognition model.Using the GGNN network to learn the hidden state embedded representation of the graph node,and then input the learned hidden state to BiLSTM-CRF for prediction.The same data set has been tested and proved that the recognition accuracy of the graph neural network model is better than BiLSTM-LAN and BiLSTM.-CRF is higher.(3)Researched the construction process of knowledge graph,including knowledge representation,knowledge extraction,knowledge storage and knowledge visualization.According to the construction process and method of general domain knowledge graph,extract medical encyclopedia semi-structured data to construct medical knowledge graph;study the visualization method of knowledge graph,use Django,JavaScript,Bootstrap,Echarts components to construct entity query and relationship query visualization based on knowledge graph Show web interface.In addition,we researched automatic question answering of knowledge graph based on rule matching,and built automatic question answering module and online medical entity recognition module.