Research On Knowledge Graph Completion Based On Graph Neural Network

In the process of the development of Internet,all kinds of information increase rapidly.It is a crucial problem to organize and express information in the process of its application.As a large-scale knowledge base for storing and organizing information,knowledge graph contains a large number of real and factual entities and various relationships between them.It was originally proposed to improve the intelligent performance of search engines,and later with its powerful Data modeling capabilities and knowledge semantic expression capabilities are widely used in intelligent question answering systems,recommendation systems,etc.,and it has become an important technology in the development of artificial intelligence.With the development of study and application based on knowledge graph and related technologies,incompleteness is a ordinary problem of the large-scale knowledge graph available now,and quiet a lot of possible entities and relationship to be mined caused by the sparsity of knowledge graph.Therefore,it is very necessary to carry out the exploration of the method of complementing the knowledge graph,and to mine or speculate the missing entity and relationship information to complement the graph structure..In this paper,a knowledge graph completion method based on graph convolution neural network and conditional random field model is proposed to integrates relational link information.This method uses the relationship link information to construct entity similarity,and uses the similarity to assist the learning of knowledge representation,and obtains the embedded expression of the graph feature.Specifically,a graph convolutional neural network and a conditional random field are used to construct an encoder,with the help of the encoder to encode the semantic information of the link relationship.Then use the decoding model to predict the new facts,automatically mine the missing information in knowledge graph,and obtain new knowledge to complete them.This method makes full use of the graph structure feature information of the knowledge graph data and the semantic information contained therein,so that the machine learning model can obtain as many effective features as possible,and then perform link prediction based on this to complete the knowledge graph.Furthermore,in order to consider the influence of neighborhood entities on the expression learning of the central entity,an attention mechanism is introduced.Based on the graph attention network and conditional random fields,a knowledge graph completion method that can integrate neighborhood structure information is proposed.Specifically,an encoder composed of a relational graph attention network and a conditional random field model is first constructed,through which the encoder captures the correlation between entities and the links between entities in different neighborhoods,and encodes the interactive features between entities and relationships,and realizes the aggregation and coding central entity and neighborhood structure information.Subsequently,the semantic holographic embedding model is used to score the predicted triples to realize link prediction,so as to realize the mining of hidden entities and relationships.In this thesis,we conducts experimental verification on the proposed models on three public knowledge graph datasets,and compares it with existing related models.Experimental results show that the model proposed in this paper performs well in link prediction tasks and can effectively improve the effect of knowledge graph completion.