Mitigating social biases of pre-trained language models via contrastive self-debiasing with double data augmentation

Keyphrases

• Data Augmentation 100%
• Debiasing 100%
• Pre-trained Language Model 100%
• Social Bias 100%
• Existing Techniques 11%
• Real-world Application 11%
• Adapter 11%
• Plug-and-play 11%
• Modeling Capability 11%
• Fairness Metrics 11%
• Language Modeling 11%
• Demographic Groups 11%
• Contrastive Learning 11%
• Augmented Data 11%
• Counterfactual Data Augmentation 11%
• Training Corpus 11%

Computer Science

• Data Augmentation 100%
• Pre-Trained Language Models 100%
• Experimental Result 11%
• World Application 11%
• Language Modeling 11%
• Bidirectional Encoder Representations From Transformers 11%
• Baseline Model 11%
• Contrastive Learning 11%
• Demographic Group 11%