Research On Knowledge Graph Link Prediction Based On Graph Neural Network

Since it proposal,the knowledge graph has developed rapidly and has shown high application value in many fields,such as assisted intelligent question answering,big data analysis,explainable artificial intelligence and so on.However,the current knowledge graph still has some shortcomings,the most prominent of which is incompleteness,that is,there are missing entities and relationships in the knowledge graph,which brings risks and challenges to the application system based on the knowledge graph.In the application of knowledge graph,link prediction is usually used to compensate for the incompleteness of knowledge graph,but the existing link prediction model has the problems of poor modeling effect on entity relations and low utilization of structural information.Therefore,this thesis uses the graph neural network model to compensate for the incompleteness of the knowledge graph and further improve the accuracy of the model,so as to achieve the knowledge graph completion.Specifically,by constructing a relational graph attention network link prediction model,high-quality relation embedding information is modeled.In addition,a link prediction model of the graph attention network combined with the relation subgraph is designed to obtain the structural feature information of the knowledge graph itself and realize the effective completion of the knowledge graph.The main research content of this thesis is divided into the following two aspects:1)A relational graph attention network link prediction model is proposed.To address the problem that the graph attention network ignores relations when aggregating entity neighborhood information,this paper effectively connects relations and head entities in the model construction to include relation information in the entity information.This improves the acquisition of relation information in the graph attention network,and the proposed model is compared with classic models such as Trans E and Conv E on public datasets such as WN18 RR and FB15K-237.Results show that the proposed model has significant experimental results,which verifies the effectiveness of the model in link prediction tasks.2)A graph attention network combined with relation subgraph structure information link prediction model is proposed.To address the shortcomings of knowledge graph link prediction models in the task of graph completion,especially the lack of information between entities and relations and structural information in the knowledge graph,this paper uses the graph attention network model to obtain semantic information of nodes and constructs a relation subgraph to obtain structural information based on relations.This enables the model to obtain more information from the knowledge graph,further improve the efficiency and accuracy of knowledge graph link prediction,and alleviate the problem of incompleteness of knowledge graph.The proposed model is verified for effectiveness on public datasets such as FB15K-237,WN18 RR,and the experimental results show that the model has the best performance on the FB15K-237 dataset and slightly weaker performance on other datasets.