Abstract: "Knowledge graph is a hot research feld in the direction of artifcial intelligence. The task of knowledge graph completion is to predict the links between entities. Translation-based models (such as TransE, TransH, and TransR) are a class of well-known knowledge graph completion methods. However, most existing translation-based models ignore the importance of triples in the completion process. In this paper, we propose a novel knowledge graph completion model PRTransE, which considers the importance information of triples based on PageRank and combines the importance information of triples with knowledge graph embedding. Specifcally, PRTransE integrates the entity importance and relationship importance of the triplet at the same time, and adopts different processing methods for the importance information of the positive and negative tuples, so that pay adaptive attention to different tripletinformation in the learning process and improve learning performance to achieve better completion effect. Experimental results show that, in two real-world knowledge graph datasets, PRTransE has the best overall performance compared to the fve comparison models."
Authors: Zhongwen Li (Harbin Institute of Technology, China); Bin Zhang (Peng Cheng Laboratory, China); Yang Liu (Harbin Institute of Technology, Shenzhen & Peng Cheng Laboratory, China); Qing Liao (Harbin Institute of Technology(Shenzhen), China)
Email: 18S151566@stu.hit.edu.cn, bin.zhang@pcl.ac.cn, liu.yang@hit.edu.cn, liaoqing@hit.edu.cn