Motif-aware Riemannian Graph Neural Networks

A new study, "Motif-aware Riemannian Graph Neural Network with Generative-Contrastive Learning," introduces MotifRGC, a novel graph neural network model that enhances Riemannian graph representation learning. The model addresses issues of curvature diversity, numerical stability, and motif regularity by using a diverse-curvature manifold and a stable kernel layer. The approach combines generative and contrastive learning to capture complex graph structures. Empirical results demonstrate its superiority over existing models in tasks like node classification and link prediction.