IterG: An Iteratively Learning Graph Convolutional Network with Ontology Semantics

Xingya Liang, Fuxiang Zhang, Xin Liu, Yajun Yang

Abstract: Knowledge reasoning aims to infer new triples based on existing triples, which is essential for the development of large knowledge graphs, especially for knowledge graph completion. With the development of neural networks, Graph Convolutional Networks (GCNs) in knowledge reasoning have been paid widespread attention in recent years. However, the GCN model only considers the structural information of knowledge graphs and ignores the ontology semantic information. In this paper, we propose a novel model named IterG, which is able to incorporate ontology semantics seamlessly into the GCN model. More specifically, IterG learns the embeddings of knowledge graphs in an unsupervised manner via GCNs and extracts the semantic ontology information via rule learning. The model is capable of propagating relation layer-wisely as well as combining both rich structural information in knowledge graphs and ontological semantics. The experimental results on five real-world datasets demonstrate that our method outperforms the state-of-the-art approaches, and IterG can effectively and efficiently fuse ontology semantics into GCNs.