GA-GGD: Improving semantic discriminability in graph contrastive learning via Generative Adversarial Network

Jitao Zhao, Dongxiao He, Meng Ge, Yongqi Huang, Lianze Shan, Yongbin Qin, Zhiyong Feng

Abstract: Highlights•A semantic embedding confusion issue is discovered in Graph Contrastive Learning.•A generative adversarial mechanism boosts semantic discriminability in GCL.•GA-GGD shows its powerful semantic discriminability in attack defense test.•GA-GGD effectiveness is proven across eight different scale datasets and three tasks.