Construction Of Knowledge Graph Oriented To Questions And Answers Of Traditional Chinese Medicine And Research On Problem Representation Algorithm

Knowledge graphs are the data foundation of question answering systems.With the rapid development of data mining and information collection technology,a large number of knowledge graphs with large scale and covering many fields are beginning to appear,but there is still a big gap in the knowledge graphs in the field of traditional Chinese medicine.Because of the particularity of Chinese medicine,the open domain knowledge graph cannot be applied to the research of TCM question and answer tasks,and the existing knowledge graph of traditional Chinese medicine cannot meet the needs of TCM question and answer tasks because of its relatively single content.Therefore,constructing a knowledge graph for TCM question answering is of great significance for improving the question answering effect of TCM question answering tasks.On the other hand,question representation is a key factor in determining the performance of question and answer tasks,and the accuracy of question representation will affect the final result of question and answer tasks.Facing complex problems,traditional text representation algorithms cannot learn the information interaction between entities and entity relationships well when learning the problem representation,mining potential text semantics,leading to the performance of question and answer tasks on complex problem processing Poor.Therefore,in this thesis,we learn the problem representation based on deep neural network to improve the question and answer effect of complex problems.The main contents of this thesis are as follows:(1)The construction of knowledge graph for TCM question and answer.In order to improve the question answering efficiency and question answering effect of TCM question answering task,this thesis completed the construction of knowledge graph suitable for TCM question answering task.This thesis completes data collection and data preprocessing based on vertical pharmaceutical websites to obtain formatted JSON files,and then implements knowledge storage through Python third-party library py2 neo to complete knowledge graph construction,which includes 7 entity types,with a scale of about 44,000 and 11 kinds of entity relationship is about 300,000.(2)Research on problem representation algorithm based on deep neural network.In order to solve the problem of poor performance of complex problems in question and answer tasks,this paper designs and implements a problem semantic graph representation model based on gated graph neural network.By displaying and modeling the structure of the semantic relationship in the problem to obtain the semantic graph of the problem and using the gated graph neural network to learn the semantic graph representation to obtain the semantic graph representation,the entities in the problem and the relationship between the entities can be better obtained Pointing.In the comparison experiment,the accuracy,recall and F1 scores of the gated graph neural network model reached: 0.2386,0.2979 and 0.2649 respectively.The experimental results show that encoding the semantic relationship structure in the question can improve the effect of complex questions in the question and answer task.