Knowledge graph completion method based on bipartite graphs and attention mechanism
To address the issue of existing knowledge graph completion methods'limited capability in capturing structural in-formation within knowledge graphs,this paper proposed a novel model that leveraged bipartite graphs and an attention mecha-nism to acquire global structural insights and facilitate automatic knowledge graph completion.This model firstly constructed two subgraphs centered on entities and relationships to capture potential useful information about entity neighborhood and rela-tionship structures,and inputted the information formed by the two subgraphs into the encoder to better update entity and rela-tionship structure information.Then,it used attention mechanisms to adaptively learn important interaction features between updated entities and relationships.Finally,it inputted the feature vectors containing global structural information into the de-coder,and it actively employed a scoring function to assess and predict scores for the input feature edges,ultimately utilizing the predicted outcomes to accomplish the task of knowledge graph completion.Comparing the performance of the proposed method with the baseline method on the FB15K-237 and NELL995 datasets,the MRR and hits@10 evaluation indicators achieved significant improvements of 5.1,8.8,and 3.4,2.2 percentage points,respectively.At the same time,on the WN18RR dataset,these two indicators also were improved by 0.1 and 1.9 percentage points,respectively.The experimental results show that established model proactively adopts a structure that effectively captures the global structural information of the knowledge graph,thereby significantly enhancing the expression ability and predictive performance of the model.
万方数据
维普
