Research On Key Technology Of Knowledge Graph Reasoning Based On Neural Network

Since the concept of knowledge graph was proposed by Google in 2012,there are lots of applications in knowledge graph,such as question answering systems,recommender systems,information retrieval systems and decision-making assistance systems.Thus,It draws great attention in the academic and industrial communities.However,the current knowledge graphs are often incomplete,sparse and facing high cost of construction,which limits their applications.How to represent and reason efficiently are still the focus and difficulty on the research of knowledge graph.In view of the low utilization and low modeling capability on the structure of the knowledge graph,we propose a joint reasoning model based on the local subgraph struc-ture and the global path structure.Besides,We extend the knowledge graph reasoning technology into the health field,and design a medical suggestion system based on the reasoning of the knowledge graph.The main work and contributions of this thesis are as follows:(1)A multi-level reasoning method based on local subgraph structure is proposed.In view of the current work's insufficient mining of the knowledge graph structure,espe-cially the multi-level relational subgraph structure,this thesis uses multi-scale convolu-tional neural networks to capture the interaction of the constructed multi-level relational subgraph elements.Finally,the effectiveness of constructing subgraph structure hierar-chically is verified through experiments.(2)A joint reasoning method combining local subgraph structure and global path structure based on attention mechanism is proposed.In view of the current model's insuf-ficient ability to model the path structure and the problem of path noise,we use CNN and Bi-LSTMs to fully capture path features.In addition,a multi-path fusion algorithm based on the entity context attention mechanism is designed.Finally the local subgraph structure and the global path structure are integrated for joint reasoning to fully excavate and utilize structure information of the knowledge graph.The experiments show the effectiveness of the path coding network,multi-path fusion strategy and joint reasoning method.(3)In terms of application,a medical assistant system based on knowledge graph reasoning is designed and implemented.The system promotes the knowledge graph rea-soning technology to the medical field closely related to personal health.In response to the problems of multi-source replacement of medical data in the medical field,the high cost of medical treatment for patients,and the lack of personal medical knowledge,a large number of text data from medical websites are integrated into the medical knowl-edge graph.Then,the medical knowledge graph can be completed in the system based on the algorithms we proposed.The system can help patients conduct self-examination of diseases,provide relevant medical advice and daily health precautions recommendations,etc.It is of practical significance for the popularization and promotion of personal medical knowledge and reducing the cost of patients' medical treatment.