Question Answering System Of Medical Knowledge Graph Based On Deep Learning

Intelligent question answering system can understand the user’s input questions and give more concise and accurate answers to the questions.It avoids the process of information filtering when users use traditional search engines,and can effectively save the time cost when users search.It has gradually become a new way of human-computer interaction in the wave of artificial intelligence.knowledge graph is a kind of storage structure in line with human cognitive form,which can accurately describe the relationship information between things,and is very suitable for providing data support for intelligent question answering.There is a huge amount of medical information on the Internet,so how to extract and use effective information has become a difficult problem.This paper selects the application of Intelligent Question Answering System in the medical field as the breakthrough point,studies the construction of medical vertical domain knowledge graph,and on this basis,builds an intelligent inquiry platform to meet the needs of users.The main research work includes the following aspects:(1)Research on the method of knowledge extraction by medical literature in Chinese based on deep learning.The purpose is to extract effective information from medical literature into knowledge.The joint extraction model of mainstream shared parameters is improved.We improved the mainstream joint extraction model based on shared parameters.By using a pre-training model based on Ro BERTa to improve the accuracy of natural language understanding,the Lattice＿LSTM neural network is adopted to replace the Bi LSTM structure to improve the adaptability of Chinese encoding task.The CRF is instead of LSTM decoding to improve the attention of the whole sentence and label constraints,and by inputting the label results of entity recognition into CNN model through shared parameters to improve the effect of relationship extraction.Finally,the Ro BERTa-Lattice＿LSTM-CRF＿CNN joint extraction model for Chinese diabetes medical literature knowledge extraction is implemented.(2)Research on how to build the knowledge graph of medical field in Chinese.The purpose is to store medical knowledge from medical literature and other sources in the form of atlas.Using crawler technology to obtain basic medical knowledge from websites such as wenyi.com and Baidu Encyclopedia;The entity alignment is carried out by calculating the text and semantic similarity which is to realize the fusion of multi-source knowledge.Combining Neo4 j and Mongo DB as relational and attribute databases to store knowledge.Finally,the Ro BERTa-Lattice＿LSTM-CRF＿CNN joint extraction model for medical literature in Chinese knowledge extraction is implemented.(3)Research on the method of knowledge graph question answering system based on semantic analysis.The purpose is to build a question and answer applications for medical knowledge graph.This part mainly studies the method of parsing question semantics in the question answering system.In order to solve the problem of cold start-up,a corpus generator is designed to obtain the training data.The entity recognition method which is based on Ro BERTa-Lattice＿LSTM-CRF and the semantic classification method which is based on Ro BERTa-Text CNN are used to interpret semantics.The answers are retrieved in the knowledge base through logical transformation.The WEB system is realized by combining Django and Layui.The graph relationship visualization is realized by using ECharts.Finally,the medical assistant consultation platform which is easy-to-use and maintainable was realized.