Automated message selection for robust Heterogeneous Graph Contrastive Learning

Rui Bing, Guan Yuan, Yanmei Zhang, Yong Zhou, Qiuyan Yan

Abstract: Highlights•Robustness of heterogeneous graph contrastive learning has not been studied yet.•Automated message selection resists the harmful messages.•The selection performs in relation and meta-path views to protect complex features.•The categorical distribution enhances the interpretability of robust learning.•A novel cross-view contrastive loss function is proposed for model optimization.