Research And Application On Knowledge Graph Representation Learning Method Based On Enhanced Information

With the rapid development of artificial intelligence,a new generation of artificial intelligence is gradually known.Cognitive intelligence,as a vital part of it,is a hot research direction of current scholars and experts.Besides,knowledge engineering is one of the cores of cognitive intelligence,as the carrier of knowledge and the cornerstone of cognitive intelligence,knowledge graph is an important data support for intelligent applications.It can realize machine language cognition and provide interpretability for artificial intelligence,it can enhance the ability of machine learning.The knowledge graph describes objective facts in the real world through a graph structure,but due to the large amount of facts,the knowledge graph is incomplete,lacking a large amount of knowledge,and many hidden knowledge is not discovered.Therefore,using knowledge graph completion to discover hidden facts and complete the knowledge graph is of great significance.Knowledge graph completion is the process of predicting new relational features based on existing knowledge in the knowledge graph,inferring implicit association information,predicting missing triplet information,and completing the knowledge graph.At present,mainstream knowledge graph completion can be summarized as follows: knowledge representation based methods,path based methods,and inference based methods,all the methods have great results.However,existing methods often only consider graph convolutional networks or first-steep neighbor node information,ignoring the graph structure information of the graph itself,and also failing to consider global relationship information.To solve the above problems,this thesis proposes a method that combines the semantic information of the knowledge map structure and the global association information to improve the completion effect of the knowledge graph.The main work of this article is as follows:(1)In response to the problem that existing graph neural networks do not consider the graph structure information of the graph itself,a graph attention network combining multi hop information is proposed.Based on the direction of the two hop relationship,the graph structure information of domain nodes and relationships is fused during node embedding to obtain node enhanced embedding;(2)To address the issue of existing graph neural networks only considering domain information,a global triple information fusion mechanism is proposed to search for other triples with the same relationship as predicted triples,aggregate global relationship information,and improve the knowledge graph completion effect;(3)Based on the above knowledge graph completion model,a medical knowledge graph intelligent question answering system has been designed and implemented,which includes knowledge graph question answering,knowledge graph completion,and node relationship query functions.This article analyzes the advantages and disadvantages of mainstream knowledge graph completion models.Based on the limitations of old models,a knowledge graph completion model based on information augmentation is proposed to improve the effectiveness of knowledge graph completion.Based on this model,an intelligent question answering system based on medical knowledge graph was designed and implemented,and various functions of the system were tested.Each function can be accurately implemented,providing convenience for users.