Temporal-structural importance weighted graph convolutional network for temporal knowledge graph completion

Haojie Nie, Xiangguo Zhao, Xin Yao, Qingling Jiang, Xin Bi, Yuliang Ma, Yongjiao Sun

Abstract: Highlights•We propose a Temporal Knowledge Graph Completion method based on temporal attention learning, named TAL-TKGC, which includes a temporal attention module and weighted GCN.•We consider the quaternions as a whole and use temporal attention to capture the deep connection between the timestamp and entities and relations at the semantic levels.•We improve the GCN which can aggregate structural information with node embedding on different weights based on the temporal semantic and structural importance of nodes.•We conducted comparison and speedup experiments on two public datasets, ICEWS14 and ICEWS05-15. The results verified the performance and the scalability on multiple GPUs of the proposed model.