Research And Implementation Of Cardiovascular Disease Question Answering System Based On Knowledge Graph

At present,the prevalence of cardiovascular disease in China is in contradiction with the construction of medical service to be perfected.With the coming of the era of "Internet Plus" and artificial intelligence,wise medical sevice has become a new direction of development.The knowledge graph becomes the cornerstone of realizing wise medical sevice,because it can extract and apply information from Massive data which has been accumulated in the medical field.Automatic question answering system based on knowledge graph can understand the user's search intent and return more accurate and effective answers.Therefore,this paper research the technologies of knowledge extraction from electronic medical records and semantic parsing based on deep learning in details,aiming at building a cardiovascular disease knowledge graph,and developing an automatic question answering system based on it,which not only provides an efficient and accurate way to access to cardiovascular disease knowledge,but also response to WIT120(Wise Information Technology of 120)policy.The main research of this article is as follows:(1)Knowledge extraction based on labeling strategy and deep learning.By improving the existing joint labeling mechanism,this article proposes a new method to mark the overlapping relationships in electronic medical records.On the basis of the mainstream extraction model BiLSTM-LSTM_Bias,introducing Adversarial trainning to improve robustness;using self-attention mechanism to fully capture statement features,adding entity decoding layer to increase the sensitivity of entity knowledge,the Adversarial-migration learning based on the shared-private domain separation network is introduced to learn the word boundary features of task sharing from other named entity recognition corpus and filter specific information,so as to improve the accuracy of the model.Finally,this paper proposes a extract model based on dual-Adversarial migration learning named JOINT-Adversarial Transfer.Experiments show that the model has a significant improvement in BiLSTM-LSTM-Bias performance,and the F1 value is improved by 4.17%.(2)The construction of the knowledge graph of cardiovascular disease.In order to enrich the knowledge graph,this article uses the packaging-based technology to extract knowledge from 39 Health Network,Baidu Encyclopedia,and completes the fusion of different sources of knowledge via related knowledge merger and entity alignment technologies.Finally,the knowledge graph of cardiovascular disease is built after the knowledge bing stored in Neo4j and other databases.(3)Automatic question answering system based on semantic parsing.In view of the fact that Word2vec can't distinguish the meaning of the same character in different contexts,this paper uses BERT to generate dynamic embedding vector based on context.What' s more,in order to reduce the labeling work,the active learning strategy based on uncertainty is used to select more enlightening samples to train model,so this paper proposes a semantic parsing scheme based on active learning and BERT.Experiments show that BERT can effectively improve the performance of the model and training methods based on active learning can achieve satisfactory performance with only 50%of the labeling corpus.Finally,an automatic question answering system of cardiovascular disease knowledge is implemented with the Vue.js framework and D3.js.