Toward General Digraph Contrastive Learning: A Dual Spatial Perspective
Researchers have introduced S2-DiGCL, a new framework designed to enhance contrastive learning for directed graphs. Unlike previous methods that primarily focused on undirected graphs, S2-DiGCL incorporates directional information crucial for real-world networks. The framework utilizes a dual spatial perspective, employing magnetic Laplacian perturbations for adaptive edge modulation and a path-based subgraph augmentation strategy to capture asymmetries. Experiments on seven real-world datasets show S2-DiGCL achieves state-of-the-art performance in node classification and link prediction. AI
IMPACT Enhances representation learning for directed graphs, potentially improving applications in social networks and recommendation systems.