NFGCL: A negative-sampling-free graph contrastive learning framework for recommendation

Yuxi Xiao, Rui Ma, Jun Sang

Abstract: Highlights•This paper provides a fresh contrastive learning framework for recommendation.•The proposed NFGCL distinguishes user preferences without negative sampling.•Proposing a novel contrastive objective, better-balancing alignment and uniformity.•Designing a representation-level augmentation method to ensure the alignment.•Experiments verify the effectiveness and superiority of the proposed method.