N-ary Relational Link Prediction Algorithm Fusing Graph Attributes

Knowledge graph is widely used in real life, but there is still a lot of missing information, which makes the completion of knowledge graph very important. Link prediction is one of the main methods to complete knowledge graph. In addition to binary relation facts which have received a lot of attention, there are also hyper-relation facts that are ubiquitous in the real world, namely n-ary relation facts. In this paper, we focus on link prediction algorithms for n-ary relation facts and find that the existing algorithms ignore the graph attribute information of nary relation facts themselves in the calculation process. Consequently, the distribution of entities and relations in n-ary relational datasets is analyzed first. The results show the fact that some of the n-ary relation facts are very important, while others are less important. This indicates that they have the characteristics of the scale-free network. Then, the global graph parameter (GGP) is introduced to describe the importance of entities and relations, and weighted to the link prediction process to improve the accuracy performance. Finally, extensive evaluation on commonly used n-ary datasets JF17K, WikiPeople, and their specific arity subsets validate the superiority of the proposed algorithm.