Research And Application Of Medical Intelligent Question Answering System Based On Knowledge Graph

With the rapid development of network information,the use of search engines such as Baidu returns a large amount of information.And a considerable amount of information is not closely related to search problems,which leads to low search efficiency and makes it difficult for users to find satisfactory answers.Therefore,search engines that can quickly and accurately provide users with the information they need have become increasingly popular,with the emergence of intelligent assistants such as Huawei Xiao E and Apple Siri.In recent years,with the continuous development of knowledge graph,knowledge graph is used to store and provide knowledge base for applications such as question and answer systems,semantic search and intelligence analysis.In the era of growing artificial intelligence,question and answer systems have attracted much attention.By using the thinking method of knowledge graph,it is possible to transform the questions asked by users into logical and clear expressions to better understand their needs.The intelligent question answering system based on knowledge graph includes many methods and technologies,and has formed a certain architecture,such as named entity recognition,relationship extraction,entity fusion,knowledge reasoning and other technologies used in the construction and storage process of knowledge graph;There are also natural language based retrieval methods in question answering systems,such as text semantic matching,intention recognition,semantic optimization methods,and semantic retrieval optimization methods.Among them,named entity recognition and text matching are key technical methods for the entire knowledge graph based intelligent question answering system.This article improves and innovates these two methods to improve the quality of the knowledge graph and the accuracy of the answer system.This paper focuses on the key technologies and applications of knowledge graph based medical intelligent Q&A,mainly including:(1)A named entity recognition method based on ERNIE2.0 pre-training model is proposed for the problem of multiple meanings of words and unbalanced label classification in Chinese named entity recognition tasks.The method increases the weight of keywords by adding a soft attention mechanism and incorporates a focal loss function to equalize the labels.The experimental results show that the method achieves better results on the named entity recognition task.(2)In view of the complex structure of Chinese sentences and the frequent occurrence of multiple meanings of a word.The fact that many current methods can only perform sentence similarity calculation by using keywords in sentences.An interrogative sentence similarity calculation method based on a combination of multi-feature fusion and twin networks is proposed.The method achieves the similarity judgment of two interrogative sentences by extracting six different feature vectors of the sentence,splicing and fusing them.Then inputting them into the twin network.A dataset of 22191 pairs of medical-related interrogative sentences was constructed by analyzing authoritative websites.This dataset was used to test the effectiveness of the present method.The experiments proved that the present method achieved good similarity matching results on the dataset.(3)A medical intelligent Q&A system based on knowledge graph is built.The knowledge graph is constructed by extracting knowledge from the authoritative medical websites certified by the Ministry of Health through crawler technology.The system focuses on four modules: issue classification,named entity recognition,template matching,translation and query.Among them,the AC algorithm is applied to the initial classification of the problem.The ERNIE2.0-Bi-LSTM-AT-CRF-FL model is applied to the named entity recognition to improve the recognition accuracy and efficiency.The word separation technique is used before entity recognition,and through template matching.The input questions can be effectively integrated with information such as question classification and entity recognition results.And compared with the templates in the question module library to generate Neo4 j query statements for knowledge graph.This paper mainly improves the named entity recognition method and sentence similarity calculation method.And completes the construction of a medical intelligent Q&A system based on knowledge graph by creating a medical knowledge graph.The experiments prove that the improved named entity recognition method and sentence similarity calculation method in this paper both have better results.After the user enters a question on the Q&A page,the system can output a more accurate answer by querying the knowledge graph and matching the translated answer.