Knowledge Graph Completion Based on Graph Neural Network

Weijun Li
School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, Ningxia, China
Haonan Li
School of Computer Science and Engineering,North Minzu University ,Yinchuan 750021, Ningxia, China
Shixia Liu
School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, Ningxia, China
Xueyang Liu
School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, Ningxia, China
Jianping Ding
School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, Ningxia, China
Xueyang Liu
School of Computer Science and Engineering, North Minzu University, Yinchuan 750021, Ningxia, China

Abstract:

Knowledge graph is a knowledge representation method that organizes entities, relationships and attributes in a graphical structure. Due to the wide application of knowledge graphs, especially in search engines, question answering systems, and recommendation systems, accurate and complete knowledge graphs have become crucial. However, the construction of a knowledge graph is a complex and time-consuming process, and a complete knowledge graph is often difficult to obtain. As a result, knowledge graph completion technique has come into being. Knowledge graph completion is a method of automatically inferring relationships and attributes between entities. Recently, a new approach based on FNN has been paid more and more attention. Graph neural networks are a class of deep learning models that have been developed to deal with graphical data. It can effectively capture the relationships and attributes between entities and perform knowledge graph completion tasks. In order to solve the problem of incomplete temporal knowledge graph, a complete algorithm TGNN based on temporal graph neural network was studied. Allocate weights to different categories of information, and introduce an attention mechanism to achieve temporal offset of event and query time data, thereby encoding special search substructures in the temporal knowledge graph. On this basis, we will study how to predict tail entities from the overall entity collection, predict tail entities in a specific small-scale entity collection, and use both to obtain the probability of tail entities. The TGNN algorithm can effectively improve the link prediction accuracy in temporal knowledge graphs.