计算机应用研究2024，Vol.41Issue(10)：2881-2892,12.DOI:10.19734/j.issn.1001-3695.2024.02.0020
针对大语言模型的偏见性研究综述
Review of biased research on large language model
摘要
Abstract
The phenomenon of bias existed widely in human society,and typically manifested through natural language.Tra-ditional bias studies have mainly focused on static word embedding models,but with the continuous evolution of natural lan-guage processing technology,research has gradually shifted towards pre-trained models with stronger contextual processing ca-pabilities.As a further development of pre-trained models,although large language mo-dels have been widely deployed in mul-tiple applications due to their remarkable performance and broad prospects,they may still capture social biases from unproc-essed training data and propagate these biases to downstream tasks.Biased large language model systems can cause adverse so-cial impacts and other potential harm.Therefore,there is an urgent need for further exploration of bias in large language mo-dels.This paper discussed the origins of bias in natural language processing and provided an analysis and summary of the deve-lopment of bias evaluation and mitigation methods from word embedding models to the current large language models,aiming to provide valuable references for future related research.关键词
自然语言处理/词嵌入/预训练模型/大型语言模型/偏见Key words
natural language processing/word embedding/pre-trained model/large language model/bias分类
信息技术与安全科学引用本文复制引用
徐磊,胡亚豪,潘志松..针对大语言模型的偏见性研究综述[J].计算机应用研究,2024,41(10):2881-2892,12.基金项目
国家自然科学基金资助项目(62076251) （62076251）
