Hyperbolic Graph Neural Networks

Graph Neural Networks (GNNs) are powerful deep representation learning methods for graphs. Most GNNs learn the node representations in Euclidean spaces. However, some studies find that compared with Euclidean geometry, hyperbolic geometry actually can provide more powerful ability to embed graphs with scale-free or hierarchical structure. As a consequence, some recent efforts begin to design GNNs in hyperbolic spaces. In this chapter, we will introduce three hyperbolic GNNs, which learn hyperbolic graph representations to get better performance.