Abstract: In recent years, significant advancements have been made in contrastive self-supervised learning for graphs. However, most of the existing methods start from the feature level and ignore the ...