Abstract: Heterogeneous graph embedding has recently advanced with two primary directions: self-supervised learning to address the scarcity of labeled data and metapath-free approaches to eliminate ...