Event-level supervised contrastive learning with back-translation augmentation for event causality identification

Shunhang Li, Gang Zhou, Jing Chen, Yepeng Sun, Ningbo Huang, Sisi Peng

Abstract: Highlights•Event causality identification based on limited and complex-labeled dataset.•Back-translation augmentation fosters causality expression diversity.•New contrastive pair sampling strategy for more causality interactions.•More robustness to unbalanced data distribution and noise.