Adaptive Multi-layer Contrastive Graph Neural Networks

Inspired by recent success of graph contrastive learning methods, we propose a self-supervised learning framework for Graph Neural Networks (GNNs) named Adaptive Multi-layer Contrastive Graph Neural Networks (AMC-GNN). Specifically, AMC-GNN generates different graph views through data augmentation and compares the output embeddings at different layers of graph neural network encoders to obtain feature representations for downstream tasks. Meanwhile, AMC-GNN learns the importance weights of embeddings at different layers adaptively through the attention mechanism, and an auxiliary encoder is adopted to train graph contrastive encoders better. The accuracy is improved by maximizing the representation's consistency of positive pairs in the intermediate layers and the final embedding space. Experiments on node classification and link prediction demonstrate that the AMC-GNN framework outperforms state-of-the-art contrastive learning methods and even sometimes outperforms supervised methods.