Preciser comparison: Augmented multi-layer dynamic contrastive strategy for text2text question classification

Jiyao Wang, Zijie Chen, Yijia Zhang, Dengbo He, Fangzhen Lin

Abstract: Highlights•A comprehensive framework (ADMC) is proposed and deployed for text2text question classification in the industry dialog system.•Multi-layered dynamic contrastive learning mechanism ensures better performance on distinguishing confusable classes.•ADMC integrates optional data augmentation, distance measuring, and loss function component.•ADMC is tested and compared with several baselines and variants on two industry datasets and three public datasets.