SORCL: Social-Reachability-driven Contrastive Learning for friend recommendation

Qiong Huang, Yahong Lian, Chunyao Song

Abstract: Highlights•First study on friend recommendations with users’ social reachability considered.•Data augmentation and adversarial learning boost embedding accuracy and robustness.•Distance-label contrastive learning enhances social reachability in recommendations.•Self-balancing multi-facet re-ranker ensures accurate and robust friend recommendations.•Experiments on three datasets validate SORCL’s superior reachability and accuracy.