PRTransE - Emphasize More Important Facts Based on Pagerank for Knowledge Graph Completion.

Zhongwen Li,Bin Zhang,Yang Liu,Qing Liao

DOI: https://doi.org/10.1007/978-3-030-59585-2_2

2020-01-01

Abstract:Knowledge graph is a hot research field in the direction of artificial 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 triplets in the completion process. In this paper, we propose a novel knowledge graph completion model PRTransE, which considers the importance information of triplets based on PageRank and combines the importance information of triplets with knowledge graph embedding. Specifically, 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 the proposed method pays adaptive attention to different triplet information 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 in terms of link prediction task compared to the five comparison models.