OAG$_{\mathrm {know}}$ know : Self-Supervised Learning for Linking Knowledge Graphs

Xiao Liu, Li Mian, Yuxiao Dong, Fanjin Zhang, Jing Zhang, Jie Tang, Peng Zhang, Jibing Gong, Kuansan Wang

Abstract: We propose a self-supervised embedding learning framework—SelfLinKG—to link concepts in heterogeneous knowledge graphs. Without any labeled data, SelfLinKG can achieve competitive performance against its supervised counterpart, and significantly outperforms state-of-the-art unsupervised methods by 26%-50% under linear classification protocol. The essential components of SelfLinKG are local attention-based encoding and momentum contrastive learning. The former aims to learn the graph representation using an attention network, while the latter is to learn a self-supervised model across knowledge graphs using contrastive learning. SelfLinKG has been deployed to build the the new version, called OAG <inline-formula><tex-math notation="LaTeX">$_{\mathrm {know}}$</tex-math></inline-formula> of Open Academic Graph (OAG). All data and codes are publicly available.