Negative sampling strategies for contrastive self-supervised learning of graph representations | Litlas