软件导刊2025，Vol.24Issue(9)：19-25,7.DOI:10.11907/rjdk.241608
预训练语言模型的偏见评估与纠正综述
A Survey of Bias Evaluation and Mitigation in Pre-trained Language Models
张展峰 1马宏伟 1姜鑫1
作者信息
- 1. 山东建筑大学 计算机科学与技术学院,山东 济南 250101
- 折叠
摘要
Abstract
The application of pre trained language models(PLMs)in the field of natural language processing is becoming increasingly wide-spread,but it also brings potential bias issues that may have unfair impacts on social groups.To this end,conduct a survey on PLMs bias as-sessment and correction.Firstly,formalize the definition of social bias in natural language processing;Secondly,summarize bias assessment datasets,methods,and correction techniques,and conduct classification discussions;Finally,point out the shortcomings in current research and future development directions.A review of bias assessment and correction related research on PLMs is beneficial for a deeper understand-ing of bias issues in PLMs and promoting the development of related correction technologies.关键词
社会偏见/偏见评估/偏见纠正/预训练语言模型/自然语言处理Key words
social bias/bias evaluation/bias mitigation/pre-trained language models/natural language processing分类
信息技术与安全科学引用本文复制引用
张展峰,马宏伟,姜鑫..预训练语言模型的偏见评估与纠正综述[J].软件导刊,2025,24(9):19-25,7.
