A Knowledge Graph Completion Method Based on Fusing Association Information
Yuhao Wang, Erping Zhao, Wei Wang · IEEE Access · 2022
Knowledge graph is a carrier of knowledge. The knowledge graph completion task is to predict the links between entities to make the knowledge graph more complete. Currently, the completion method based on knowledge representation learning ignores the association information between each relation in the multiple-step relation paths and the direct relations, and the association information between the head and tail entity types and direct relations. In this study, we extract and utilize these associated information, and propos AiTransE model for knowledge graph completion, which uses the frequency of each relation in the multiple-step relation paths between head and tail entities to calculate the degree of association with the direct relations, and uses head and tail entity types and direct relation types for matching to obtain the degree of association between them, Finally, the two association degrees are linearly weighted and merged and then introduced into the objective function, so that the model can give different degrees of attention to different triples, and improve model knowledge representation learning performance. The experiment with link prediction was carried out on the Tibet Animal Husbandry Dataset and the WN18 dataset, and compared with the TransE, TransH, TransR, and other models. The experimental results show that this model has a significant improvement over other models in the indicators Hits@10 and Mean Rank.