Self-supervised graph representation learning via bootstrapping

Graph neural networks (GNNs) apply deep learning techniques to graph-structured data and have achieved promising performance in graph representation learning. However, existing GNNs rely heavily on labeled or well-designed negative samples. To address these issues, we propose a new self-supervised method: bootstrapping (DGB). DGB consists of two networks: online target networks, the input them are different augmented views initial graph. The network is trained predict while updated with slow-moving average network, which means can learn from each other. As result, proposed without examples an unsupervised manner. In addition, summarize three kinds augmentation methods for DGB. Experiments benchmark datasets show performs better than current state-of-the-art how affect performances.