Towards fair decision: A novel representation method for debiasing pre-trained models

Junheng He, Nankai Lin, Qifeng Bai, Haoyu Liang, Dong Zhou, Aimin Yang

Abstract: Highlights•We propose PCFR that utilizes prompt and contrastive learning to debias information.•Our method refines sentence interactions to emphasize intricate semantic details.•PCFR effectively debiases and can greatly preserves the performance.