How do Graph Neural Networks enable unsupervised representation learning?

GNNs learn node embeddings by aggregating neighborhood information without labels.

Graph Neural Networks propagate information across nodes using message passing. In unsupervised settings, objectives like DeepWalk, node2vec, or contrastive graph learning (e.g., DGI, GraphCL) are used. These methods optimize embeddings such that structurally similar nodes have similar representations, often using random walks or mutual information maximization.