多模态大模型的偏见溯源和危害以及公平性提升研究综述

陈晋音; 王泽楷; 沈文杰; 张甜馨; 郑海斌; 邱洪君

doi:10.11959/j.issn.2096-109x.AQ25092

您当前的位置：

首页 >

文章列表页 >

多模态大模型的偏见溯源和危害以及公平性提升研究综述

综述 | 更新时间：2026-05-11

- 多模态大模型的偏见溯源和危害以及公平性提升研究综述
- Survey on multimodal large language model: bias origins, bias harms, and fairness enhancement
- 网络与信息安全学报 2026年12卷第2期页码：22-40
- 作者机构：
  
  1.浙江工业大学计算机科学与技术学院，浙江杭州 310023
  2.浙江工业大学信息工程学院，浙江杭州 310023
  3.四川大学数据安全防护与智能治理教育部重点实验室，四川成都 610065
  4.杭州电子科技大学网络空间安全学院，浙江杭州 310018
- 作者简介：
  
  [ "陈晋音（1982- ），女，浙江工业大学教授，主要研究方向为人工智能安全、图数据挖掘和进化计算。" ]
  [ "王泽楷（2004- ），男，浙江工业大学本科生。" ]
  [ "沈文杰（2002- ），男，浙江工业大学硕士生，主要研究方向为深度学习和人工智能安全。" ]
  [ "张甜馨（2002- ），女，浙江工业大学硕士生，主要研究方向为深度学习和人工智能安全。" ]
  [ "郑海斌（1995- ），男，浙江工业大学助理研究员，主要研究方向为深度学习和人工智能安全。" ]
  [ "邱洪君（1982- ），女，博士，杭州电子科技大学网络空间安全学院讲师，主要研究方向为人工智能安全、工业互联网安全。" ]
- 基金信息：
  
  国家自然科学基金资助项目(62406286;62072406);浙江省自然科学基金(LDQ23F020001);四川大学数据安全防护与智能治理教育部重点实验室开放课题(SCUSAKFKT202402Z);北京生命科技研究院有限公司开放基金资助项目(2024200CD0210)
- DOI：10.11959/j.issn.2096-109x.AQ25092
  中图分类号： TP393
- 收稿：2025-11-26，
  
  修回：2026-03-19，
  
  录用：2026-03-20，
  
  纸质出版：2026-04-25
- 稿件说明：
移动端阅览
陈晋音, 王泽楷, 沈文杰, 等. 多模态大模型的偏见溯源和危害以及公平性提升研究综述[J]. 网络与信息安全学报, 2026, 12(2): 22-40.

Chen Jinyin, Wang Zekai, Shen Wenjie, et al. Survey on multimodal large language model: bias origins, bias harms, and fairness enhancement[J]. Chinese Journal of Network and Information Security, 2026, 12(2): 22-40.
陈晋音, 王泽楷, 沈文杰, 等. 多模态大模型的偏见溯源和危害以及公平性提升研究综述[J]. 网络与信息安全学报, 2026, 12(2): 22-40. DOI： 10.11959/j.issn.2096-109x.AQ25092.

Chen Jinyin, Wang Zekai, Shen Wenjie, et al. Survey on multimodal large language model: bias origins, bias harms, and fairness enhancement[J]. Chinese Journal of Network and Information Security, 2026, 12(2): 22-40. DOI： 10.11959/j.issn.2096-109x.AQ25092.

摘要

以ChatGPT为代表的大语言模型（large language model

LLM）的快速发展，对人类社会的影响力日益凸显。随着LLM的广泛应用，其潜在的安全漏洞不断暴露，引发学术界和产业界的高度关注，也使得安全评估等相关方向成为研究热点。为此，基于多模态大模型（multimodal large language model

MLLM）的视角，对MLLM的偏见问题与安全风险进行了系统性分析。首先，明确界定了MLLM偏见的定义，深入剖析了模型偏见的三大来源（即训练数据、模型设计缺陷和训练过程）。其次，从3个维度（基于嵌入的评估方法、基于概率的评估方法和基于生成文本的评估方法）对MLLM偏见评估方法进行系统梳理。在此基础上，对近5年的偏见缓解研究成果进行了分类梳理。最后，探讨了MLLM偏见缓解面临的机遇与挑战，以期为研究人员、从业者及政策制定者的MLLM应用与研究工作提供参考。

Abstract

The rapid development of large language model (LLM)

represented by ChatGPT

has exerted an increasingly profound impact on human society. With the widespread deployment of LLM in real-world applications

their potential security vulnerabilities and social risks have been continuously exposed

attracting growing attention from both academia and industry and making safety evaluation an important research topic. From the perspective of multimodal large language model (MLLM)

this survey provides a systematic analysis of bias issues and security risks in MLLM. First

the definition of bias in MLLM is clarified

and three major sources of model bias are examined in depth

including training data

model design deficiencies

and the training process. Second

existing bias evaluation methods for MLLM are systematically reviewed from three dimensions: embedding-based evaluation

probability-based evaluation

and generated-text-based evaluation. On this basis

recent advances in bias mitigation over the past five years are categorized and summarized. Finally

the opportunities and challenges of bias mitigation in MLLM are discussed

with the aim of providing useful references for researchers

practitioners

and policymakers in the application and research of MLLM.

关键词

Keywords

references

Islam M R . Application of multimodal large language models in autonomous driving [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2412 . 16410 .

Ye Q L , Yu Z T , Shao R , et al . CAT: enhancing multimodal large language model to answer questions in dynamic audio-visual scenarios [C ] // Proceedings of the 18th European Conference Computer Vision-ECCV 2024 . Berlin : Springer , 2024 : 146 - 164 .

Liu F L , Li Z , Yin Q Y , et al . A multimodal multidomain multilingual medical foundation model for zero shot clinical diagnosis [J ] . NPJ Digital Medicine , 2025 , 8 ( 1 ): 86 .

Bai Z C , Wang P C , Xiao T J , et al . Hallucination of multimodal large language models: a survey [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2404. 1893 .

Li H R , Chen Y L , Luo J L , et al . Privacy in large language models: attacks, defenses and future directions [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2310.10383.

Maity K , Poornash A S , Saha S , et al . ToxVidLM: a multimodal framework for toxicity detection in code-mixed videos [C ] // Proceedings of the Findings of the Association for Computational Linguistics ACL 2024 . Stroudsburg : ACL , 2024 : 11130 - 11142 .

Chen M , Cao Y , Zhang Y , et al . Quantifying and mitigating unimodal biases in multimodal large language models: a causal perspective [C ] // Proceedings of the Findings of the Association for Computational Linguistics: EMNLP 2024 . Stroudsburg : ACL , 2024 : 16449 - 16469 .

Adewumi T , Alkhaled L , Gurung N , et al . Fairness and bias in multi‐modal AI: a survey [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2406 . 19097 .

Vanmassenhove E , Emmery C , Shterionov D . NeuTral rewriter: a rule-based and neural approach to automatic rewriting into gender neutral alternatives [C ] // Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing . Stroudsburg, PA, USA : ACL , 2021 : 8940 - 8948 .

Bansal H , Yin D , Monajatipoor M , et al . How well can text-to-image generative models understand ethical natural language inter‐ventions? [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2210.15230.

Kumar S , Balachandran V , Njoo L , et al . Language generation models can cause harm: so what can we do about it? an actionable survey [C ] // Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics . Stroudsburg : ACL , 2023 : 3299 - 3321 .

Yu T , Zhang Y F , Fu C Y , et al . Aligning multimodal LLM with human preference: a survey [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2503 . 14504 .

Gulati A , Lepri B , Oliver N . Lookism: the overlooked bias in computer vision [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2408.11448.

Narnaware V , Vayani A , Gupta R , et al . SB-Bench: stereotype bias benchmark for large multimodal models [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2502 . 08779 .

Bavalatti T , Ahmed O , Potdar D , et al . A systematic review of open datasets used in text-to-image (T2I) gen AI model safety [J ] . IEEE Access , 2025 , 13 : 32661 - 32680 .

Peng B L , Chen K Y , Li M , et al . Securing large language models: addressing bias, misinformation, and prompt attacks [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2409.08087.

Mehrabi N , Morstatter F , Saxena N , et al . A survey on bias and fairness in machine learning [J ] . ACM Computing Surveys , 2022 , 54 ( 6 ): 1 - 35 .

Gupta V , Venkit P N , Wilson S , et al . Sociodemographic bias in language models: a survey and forward path [C ] // Proceedings of the 5th Workshop on Gender Bias in Natural Language Processing(GeBNLP) . Stroudsburg : ACL , 2024 : 295 - 322 .

Ferrara E . Should ChatGPT be biased? Challenges and risks of bias in large language models [J ] . First Monday , 2023 , 28 ( 11 ): 13346 .

Zhang Z X , Song J . Developments in the studies of prejudice [J ] . Psychology and Behavior Research , 2007 , 5 ( 2 ): 150 - 155 .

Jussim L , Honeycutt N . Bias in psychology: a critical, historical and empirical review [J ] . Swiss Psychology Open: the Official Journal of the Swiss Psychological Society , 2024 , 4 ( 1 ): 5 .

Gordon W A . The nature of prejudice [M ] . Reading, MA : Addison-Wesley , 1954 .

Bolukbasi T , Chang K W , Zou J Y , et al . Man is to computer programmer as woman is to homemaker? debiasing word embeddings [C ] // Proceedings of the Advances in Neural Information Processing Systems 29: Annual Conference Advances in Neural Information . San Diego : Neural Information Processing Systems , 2016 : 4349 - 4357 .

Kumar A , Yunusov S , Emami A . Subtle biases need subtler measures: dual metrics for evaluating representative and affinity bias in large language models [C ] // Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics . Stroudsburg : ACL , 2024 : 375 - 392 .

Alvarez J M , Colmenarejo A B , Elobaid A , et al . Policy advice and best practices on bias and fairness in AI [J ] . Ethics and Information Technology , 2024 , 26 ( 2 ): 31 .

Ferrer X , Van Nuenen T , Such J M , et al . Bias and discrimination in AI: a cross-disciplinary perspective [J ] . IEEE Technology and Society Magazine , 2021 , 40 ( 2 ): 72 - 80 .

Ferrara E . Fairness and bias in artificial intelligence: a brief survey of sources, impacts, and mitigation strategies [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2304 . 07683 .

Glickman M , Sharot T . How human–AI feedback loops alter human perceptual, emotional and social judgements [J ] . Nature Human Behaviour , 2024 , 9 ( 2 ): 345 - 359 .

Messeri L , Crockett M J . Artificial intelligence and illusions of understanding in scientific research [J ] . Nature , 2024 , 627 : 49 - 58 .

Gallegos I O , Rossi R A , Barrow J , et al . Bias and fairness in large language models: a survey [J ] . Computational Linguistics , 2024 , 50 ( 3 ): 1097 - 1179 .

Delobelle P , Tokpo E K , Calders T , et al . Measuring fairness with biased rulers: a comparative study on bias metrics for pre-trained language models [C ] // Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Lin-guistics: Human Language Technologies . Stroudsburg : ACL , 2022 : 1693 - 1706 .

Franco D , Navarin N , Donini M , et al . Deep fair models for complex data: graphs labeling and explainable face recognition [J ] . Neurocomputing , 2022 , 470 : 318 - 334 .

Liang P P , WU C Y , Morency L P , et al . Towards understanding and mitigating social biases in language models [C ] // Proceedings of the International Conference on Machine Learning . London : PMLR , 2022 : 6565 - 6576 .

Zhang X Y , Zhai J , Ma S Q , et al . Unveiling provider bias in large language models for code generation [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2501.07849.

Blodgett S L , Barocas S , Daumé III H , et al . Language (technology) is power: a critical survey of “bias” in NLP [C ] // Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics . Stroudsburg : ACL , 2020 : 5454 - 5476 .

Tang S C , Zhu H M . Mitigating bias in generative AI: a comprehensive framework for governance and accountability [J ] . Law , Ethics & Technology, 2024 , 1 ( 3 ): 19 - 20 .

Meade N , Poole-Dayan E , Reddy S . An empirical survey of the effectiveness of debiasing techniques for pre-trained language models [C ] // Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics . Stroudsburg : ACL , 2022 : 1878 - 1898 .

Li Y J , Du M N , Song R , et al . A survey on fairness in large language models [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2308.10149.

Lin Z C , Guan S Y , Zhang W D , et al . Towards trustworthy LLMs: a review on debiasing and dehallucinating in large language models [J ] . Artificial Intelligence Review , 2024 , 57 ( 9 ): 243 .

Cheng H , Guo Y , Guo Q , et al . Social debiasing for fair multi-modal LLMs [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2408.06569.

Birhane A , Dehdashtian S , Prabhu V , et al . The dark side of dataset scaling: evaluating racial classification in multimodal models [C ] // Proceedings of the 2024 ACM Conference on Fairness, Account ability, and Transparency . New York : ACM , 2024 : 1229 - 1244 .

Garcia N , Hirota Y , Wu Y K , et al . Uncurated image-text datasets: shedding light on demographic bias [C ] // Proceedings of the 2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) . Piscataway : IEEE Press , 2023 : 6957 - 6966 .

Haliburton L , Leusmann J , Welsch R , et al . Uncovering labeler bias in machine learning annotation tasks [J ] . AI and Ethics , 2025 , 5 ( 3 ): 2515 - 2528 .

Hovy D , Prabhumoye S . Five sources of bias in natural language processing [J ] . Language and Linguistics Compass , 2021 , 15 ( 8 ): e12432 .

Mavrogiorgos K , Kiourtis A , Mavrogiorgou A , et al . Bias in machine learning: a literature review [J ] . Applied Sciences , 2024 , 14 ( 19 ): 8860 .

Hirota Y , Nakashima Y , Garcia N . Gender and racial bias in visual question answering datasets [C ] // Proceedings of the 2022 ACM Conference on Fairness Accountability and Transparency . New York : ACM , 2022 : 1280 - 1292 .

Khan M J , Breslin J G , Curry E . Towards fairness in multimodal scene graph generation: mitigating biases in datasets, knowledge sources and models [C ] // MUWS 2023: Proceedings of the 2nd International Workshop on Multimodal Human Understanding for the Web and Social Media . New York : ACM , 2023 : 34 - 46 .

Park Y , Woo S , Lee S , et al . Cross-modal alignment and translation for missing modality action recognition [J ] . Computer Vision and Image Understanding , 2023 , 236 : 103805 .

Pan Z , Jiang S , Yang X , et al . Hierarchical cross-modal image generation for multimodal biometric recognition with missing modality [J ] . IEEE Transactions on Information Forensics and Security , 2025 , 20 : 4308 - 4321 .

Yin S K , Fu C Y , Zhao S R , et al . A survey on multimodal large language models [J ] . National Science Review , 2024 , 11 ( 12 ): nwae403 .

Zhang Y F , Yu W , Wen Q , et al . Debiasing multimodal large language models via penalization of language priors [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2403 . 05262 .

Shah M , Sureja N . A comprehensive review of bias in deep learning models: methods, impacts, and future directions [J ] . Archives of Computational Methods in Engineering , 2025 , 32 ( 1 ): 255 - 267 .

Hickman L , Huynh C , Gass J , et al . Whither bias goes, I will go: an integrative, systematic review of algorithmic bias mitigation [J ] . Journal of Applied Psychology , 2025 , 110 ( 7 ): 979 - 1000 .

Raj C , Mukherjee A , Caliskan A , et al . Breaking bias, building bridges: evaluation and mitigation of social biases in LLMs via contact hypothesis [J ] . ACM Conference on AI, Ethics, and Society , 2024 , 7 : 1180 - 1189 .

Xiao X , Wu B H , Wang J C , et al . Seeing the image: prioritizing visual correlation by contrastive alignment [C ] // Proceedings of the Advances in Neural Information Processing Systems 37 . San Diego : Neural Information Processing Systems , 2024 : 30925 - 30950 .

Hamidieh K , Zhang H , Gerych W , et al . Identifying implicit social biases in vision-language models [C ] // Proceedings of the AAAI/ ACM Conference on AI , Ethics, and Society . New York : ACM Press , 2024 : 547 - 561 .

Gautam S , Srinath M . Blind spots and biases: exploring the role of annotator cognitive biases in NLP [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2404 . 19071 .

Wang Y , Yin Y X , Li P . Towards the mitigation of confirmation bias in semi-supervised learning: a debiased training perspective [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 22409 . 18316 .

Wolfe R , Yang Y W , Howe B , et al . Contrastive language-vision AI models pretrained on web-scraped multimodal data exhibit sexual objectification bias [C ] // Proceedings of the 2023 ACM Conference on Fairness Accountability and Transparency . New York : ACM , 2023 : 1174 - 1185 .

Leng S , Xing Y , Cheng Z , et al . The curse of multi-modalities: evaluating hallucinations of large multimodal models across language, visual, and audio [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2410 . 12787 .

Seyyed-Kalantari L , Zhang H R , Mcdermott M B A , et al . Underdiagnosis bias of artificial intelligence algorithms applied to chest radiographs in under-served patient populations [J ] . Nature Medicine , 2021 , 27 ( 12 ): 2176 - 2182 .

Farnadi G , Havaei M , Rostamzadeh N . Position: cracking the code of cascading disparity towards marginalized communities [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2406 . 01757 .

Xu C J , Zhang J W , Chen Z R , et al . MMDT: decoding the trustworthiness and safety of multimodal foundation models [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2503 . 14827 .

Wang Y Y , Wang X Z , Beutel A , et al . Understanding and improving fairness-accuracy trade-offs in multi-task learning [C ] // Proceedings of the 27th ACM SIGKDD Conference on Knowledge Discovery & Data Mining . New York : ACM , 2021 : 1748 - 1757 .

Hu F , Ratz P , Charpentier A . Fairness in multi-task learning via Wasserstein barycenters [M ] // Machine Learning and Knowledge Discovery in Databases: Research Track . Cham : Springer Nature Switzerland , 2023 : 295 - 312 .

Aguirre C , Dredze M . Transferring fairness using multi-task learning with limited demographic information [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2305 . 12671 .

Li X , Chen Z , Zhang J M , et al . Benchmarking bias in large language models during role-playing [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2411 . 00585 .

Zhuo T Y , Huang Y J , Chen C Y , et al . Red teaming ChatGPT via jailbreaking: bias, robustness, reliability and toxicity [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2301 . 12867 .

Khandelwal K , Tonneau M , Bean A M , et al . Indian-BhED: a dataset for measuring India-centric biases in large language models [C ] // Proceedings of the 2024 International Conference on Information Technology for Social Good . New York : ACM , 2024 : 231 - 239 .

Wang B X , Chen W X , Li B . DecodingTrust: a comprehensive assessment of trustworthiness in GPT models [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2306 . 11698 .

Akhras N , Antaki F , Mottet F , et al . Large language models perpetuate bias in palliative care: development and analysis of the palliative care adversarial dataset (PCAD) [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2502 . 08073 .

Rutinowski J , Franke S , Endendyk J , et al . The self-perception and political biases of ChatGPT [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2304.07333.

Hartmann J , Schwenzow J , Witte M . The political ideology of conversational AI: converging evidence on ChatGPT’s proenvironmental, left-libertarian orientation [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2301.01768.

Feng S B , Park C Y , Liu Y H , et al . From pretraining data to language models to downstream tasks: tracking the trails of political biases leading to unfair NLP models [C ] // Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) . Stroudsburg : ACL , 2023 : 11737 - 11762 .

Bachmann D , Van Der Wal O , Chvojka E , et al . Ffl-IRT-ing with psychometrics to improve NLP bias measurement [J ] . Minds and Machines , 2024 , 34 ( 4 ): 37 .

Zhang T , Zeng Z Q , Xiao Y X , et al . GenderAlign: an alignment dataset for mitigating gender bias in large language models [C ] // Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics . Stroudsburg : ACL , 2025 : 11293 - 11311 .

Selvam N R , Dev S , Khashabi D , et al . The tail wagging the dog: dataset construction biases of social bias benchmarks [PP ] . arXiv [ 2026-03-24 ] . arXiv:ar Xiv. 2210 . 10040 .

Dhamala J , Sun T , Kumar V , et al . BOLD: dataset and metrics for measuring biases in open-ended language generation [C ] // Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency . New York : ACM , 2021 : 862 - 872 .

Bevara R V K , Mannuru N R , Karedla S P , et al . Scaling implicit bias analysis across transformer-based language models through embedding association test and prompt engineering [J ] . Applied Sciences , 2024 , 14 ( 8 ): 3483 .

Loon A V , Giorgi S , Willer R , et al . Regional negative bias in word embeddings predicts racial animus: but only via name frequency [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2201 . 08451 .

May C , Wang A , Bordia S , et al . On measuring social biases in sentence encoders [C ] // Proceedings of the 2019 Conference of the North . Stroudsburg : ACL , 2019 : 622 - 628 .

Gray M , Milanova M , Wu L . Enhancing bias assessment for complex term groups in language embedding models: quantitative comparison of methods [J ] . JMIR Medical Informatics , 2024 , 12 : e60272 .

Guo W , Caliskan A . Detecting emergent intersectional biases: contextualized word embeddings contain a distribution of human-like biases [C ] // Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society . New York : ACM , 2021 : 122 - 133 .

Dolci T , Azzalini F , Tanelli M . Improving gender-related fairness in sentence encoders: a semantics-based approach [J ] . Data Science and Engineering , 2023 , 8 ( 2 ): 177 - 195 .

Lucy L , Bamman D . Gender and representation bias in GPT-3 generated stories [C ] // Proceedings of the Third Workshop on Narrative Understanding . Stroudsburg : ACL , 2021 : 48 - 55 .

Marinucci L , Mazzuca C , Gangemi A . Exposing implicit biases and stereotypes in human and artificial intelligence: state of the art and challenges with a focus on gender [J ] . AI & Society , 2023 , 38 ( 2 ): 747 - 761 .

Nadeem M , Bethke A , Reddy S . StereoSet: measuring stereotypical bias in pretrained language models [C ] // Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing . Stroudsburg : ACL , 2021 : 5356 - 5371 .

Nangia N , Vania C , Bhalerao R , et al . CrowS-pairs: a challenge dataset for measuring social biases in masked language models [C ] // Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) . Stroudsburg : ACL , 2020 : 1953 - 1967 .

Kaneko M , Bollegala D , Okazaki N . Debiasing isn't enough!on the effectiveness of febiasing MLMs and their social biases in downstream tasks [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2309 . 07251 .

Barikeri S , Lauscher A , Vulić I , et al . RedditBias: a real-world resource for bias evaluation and debiasing of conversational language models [C ] // Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing . Stroudsburg : ACL , 2021 : 1941 - 1955 .

Bommasani R , Liang P , Lee T . Holistic evaluation of language models [J ] . Annals of the New York Academy of Sciences , 2023 , 1525 ( 1 ): 140 - 146 .

Cheng M , Durmus E , Jurafsky D . Marked personas: using natural language prompts to measure stereotypes in language models [C ] // Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics . Stroudsburg : ACL , 2023 : 1504 - 1532 .

Omrani S S , Wolfe R , Caliskan A . Evaluating biased attitude associations of language models in an intersectional context [C ] // Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society . New York : ACM , 2023 : 542 - 553 .

Alnegheimish S , Guo A , Sun Y . Using natural sentence prompts for understanding biases in language models [C ] // Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies . Stroudsburg : ACL , 2022 : 2824 - 2830 .

An H Z , Li Z X , Zhao J Y , et al . SODAPOP: open-ended discovery of social biases in social commonsense reasoning models [C ] // Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics . Stroudsburg : ACL , 2022 : 1565 - 1588 .

Garimella A , Amarnath A , Kumar K , et al . He is very intelligent, she is very beautiful? on mitigating social biases in language modelling and generation [C ] // Proceedings of the Findings of the Association for Computational Linguistics: ACL-IJCNLP 2021 . Stroudsburg : ACL , 2021 : 4534 - 4545 .

Garimella A , Mihalcea R , Amarnath A . Demographic-aware language model fine-tuning as a bias mitigation technique [C ] // Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Processing . Stroudsburg : ACL , 2022 : 311 - 319 .

Zhou G Y , Yan Y B , Zou X , et al . Mitigating modality prior-induced hallucinations in multimodal large language models via deciphering attention causality [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2410 . 04780 .

Qian R , Ross C , Fernandes J , et al . Perturbation augmentation for fairer NLP [C ] // Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing . Stroudsburg : ACL , 2022 : 9496 - 9521 .

Theodorou B , Danek B , Tummala V , et al . Improving medical machine learning models with generative balancing for equity and excellence [J ] . NPJ Digital Medicine , 2025 , 8 ( 1 ): 100 .

Penedo G , Malartic Q , Hesslow D , et al . The refinedWeb dataset for Falcon LLM: outperforming curated corpora with web data, and web data only [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2306 . 01116 .

Venkit P N , Gautam S , Panchanadikar R , et al . Nationality bias in text generation [C ] // Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics . Stroudsburg : ACL , 2023 : 116 - 123 .

Sun T , Webster K , Shah A , et al . They, them, theirs: rewriting with gender-neutral English [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2102.06788.

Lauscher A , Lueken T , Glavaš G . Sustainable modular debiasing of language models [C ] // Proceedings of the Conference on Empirical Methods in Natural Language Processing . Stroudsburg : ACL , 2021 : 4782 - 4797 .

Yang K , Yu C , Fung Y R , et al . ADEPT: a debiasing prompt framework [C ] // Proceedings of the AAAI Conference on Artificial Intelligence . Vancouver : AAAI Press , 2023 : 10780 - 10788 .

Wang R , Cheng P , Henao R . Toward fairness in text generation via mutual information minimization based on importance sampling [C ] // Proceedings of the International Conference on Artificial Intelligence and Statistics . London : PMLR , 2023 : 4473 - 4485 .

Tian J J , Emerson D , Miyandoab S Z , et al . Soft-prompt tuning for large language models to evaluate bias [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2306 . 04735 .

Wang X , Ge T , Mao A , et al . Pay attention to your tone: introducing a new dataset for polite language rewrite [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2212.10190.

Majumder B , He Z X , Mcauley J J . Interfair: debiasing with natural language feedback for fair interpretable predictions [C ] // Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing . Stroudsburg : ACL , 2022 : 9466 - 9471 .

Amrhein C , Schottmann F , Sennrich R , et al . Exploiting biased models to de-bias text: a gender-fair rewriting model [C ] // Proceed‐ings of the 61st Annual Meeting of the Association for Computational Linguistics . Stroudsburg : ACL , 2023 : 4486 - 4506 .

Tokpo E K , Calders T . Text style transfer for bias mitigation using masked language modeling [C ] // Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Student Research Workshop . Stroudsburg : ACL , 2022 : 163 - 171 .

Alhafni B , Habash N , Bouamor H . User-centric gender rewriting [C ] // Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies: Student Research Workshop . Stroudsburg : ACL , 2022 : 618 - 631 .

Jain N , Popovic M , Groves D , et al . Generating gender augmented data for NLP [PP ] . arXiv [ 2026-03-24 ] . arXiv: arXiv. 2107.05987.

Ruan X W , Qi Z B , Xu Y R , et al . Dual-guided multi-modal bias removal strategy for temporal sentence grounding in video [J ] . Multimedia Systems , 2025 , 31 ( 2 ): 108 .

Janghorbani S , De Melo G D . Multimodal bias: introducing a framework for stereotypical bias assessment beyond gender and race in vision language models [PP ] . arXiv [ 2026-03-24 ] . arXiv:arXiv. 2303.12734.

Yang D K , Li M C , Xiao D L , et al . Towards multimodal sentiment analysis debiasing via bias purification [C ] // Proceedings of the European Conference on Computer Vision . Cham : Springer Nature Switzerland , 2024 : 464 - 481 .

Hastings B C , Qian L J , Gibson C , et al . Data augmentation via diffusion model to enhance AI fairness [J ] . Frontiers in Artificial Intelligence , 2025 , 8 : 1530397 .

Poncibò C , Ebers M . Comparative perspectives on the regulation of large language models [J ] . Cambridge Forum on AI: Law and Governance . Cambridge University Press , 2025 , 1 : e10 .

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

暂无数据