Library and Information Service

Library and Information Service >

2026 , Vol. 70 >Issue 8: 98 - 112

DOI: https://doi.org/10.13266/j.issn.0252-3116.2026.08.008

RESEARCH PAPERS

Evaluating Fairness Recognition Capabilities in Chinese Large Language Models: Construction of a Dataset and Analysis of Model Performance

  • Zhu Shucheng ,
  • Wang Weikang ,
  • Liu Ying ,
  • Liu Pengyuan
Expand
  • 1 Libraries, Renmin University of China, Beijing 100872;
    2 School of Humanities, Tsinghua University, Beijing 100084;
    3 School of Information Management and Engineering, Shanghai University of Finance and Economics, Shanghai 200433;
    4 College of Information Science, Beijing Language and Culture University, Beijing 100083;
    5 Language Resources Monitoring and Research Center Print Media Language Branch, Beijing Language and Culture University, Beijing 100083
Zhu Shucheng, PhD;Wang Weikang, Doctoral Candidate;Liu Ying, professor, PhD, Corresponding Author, E-mail: yingliu@tsinghua.edu.cn;Liu Pengyuan, Senior Research Fellow, PhD.

Received date: 2025-01-22

  Revised date: 2025-05-31

  Online published: 2026-04-08

Supported by

This work is supported by the CCF-Baidu Open Fund project titled “Construction of Fairness Datasets and Evaluation Standards for Large Language Models” (Grant No. CCF-BAIDU 202323) and National Language Commission Foundation of China “Resource and Platform Development for Measuring Bias in Linguistic Behavior of Large Language Models” (ZDI145-97).

Fold

Abstract

[Purpose/Significance] This paper evaluates the capability of Chinese large language models to recognize textual fairness, constructs a corresponding fairness evaluation dataset and methodological frameworks, and promotes the fair and healthy development of large language models. [Method/Process] This paper constructed a sentence-level multidimensional fairness dataset comprising ten group categories, five text genres, and three fairness recognition tasks. The dataset was designed to evaluate the fairness recognition capabilities of eleven language models, including neural network language models, pre-trained language models, and large language models. Additionally, it examined changes in the performance of large language models after in-context learning. [Result/Conclusion] Compared to the relatively objective task of sentiment recognition, large language models exhibit a decline in performance on the more subjective task of harmfulness recognition, especially when considering the potential for harmfulness. The models struggle to learn new human-assigned values solely through instruction from prompt-based directives. In both sentiment recognition and harmfulness recognition tasks, large language models underperform compared to fine-tuned pre-trained language models specifically designed for these tasks. Additionally, the fairness recognition capabilities of large language models are influenced by the text's genre. However, after in-context learning through prompts, the fairness recognition capabilities of large language models show notable improvement.

Key words： large language models; fairness; evaluation; dataset

Cite this article

Zhu Shucheng , Wang Weikang , Liu Ying , Liu Pengyuan . Evaluating Fairness Recognition Capabilities in Chinese Large Language Models: Construction of a Dataset and Analysis of Model Performance[J]. Library and Information Service, 2026 , 70(8) : 98 -112 . DOI: 10.13266/j.issn.0252-3116.2026.08.008

References

[1] STOKEL-WALKER C, VAN NOORDEN R. What ChatGPT and generative AI mean for science[J]. Nature, 2023, 614(7947):214-216.
[2] 王昱颖,张敏,杨晶然,等.深度学习模型中的公平性研究[J].软件学报, 2023, 34(9):4037-4055.(WANG Y Y, ZHANG M, YANG J R, et al. Research on fairness in deep learning models[J]. Journal of software, 2023, 34(9):4037-4055.)
[3] SALINAS A, SHAH P, HUANG Y, et al. The unequal opportunities of large language models:examining demographic biases in job recommendations by ChatGPT and Llama[C] //Proceedings of the 3rd ACM conference on equity and access in algorithms, mechanisms, and optimization. Boston:ACM, 2023:1-15.
[4] BREITFELLER L, AHN E, JURGENS D, et al. Finding microaggressions in the wild:a case for locating elusive phenomena in social media Posts[C] //Proceedings of the 2019conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing. Hong Kong:ACL, 2019:1664-1674.
[5] HADA R, SETH A, DIDDEE H, et al.“Fifty shades of bias”:normative ratings of gender bias in GPT generated English text[C] //Proceedings of the 2023 conference on empirical methods in natural language processing. Singapore:ACL, 2023:1862-1876.
[6] DESHPANDE A, MURAHARI V, RAIJPUROHIT T, et al.Toxicity in ChatGPT:analyzing persona-assigned language models[C] //Findings of the association for computational linguistics:EMNLP 2023. Singapore:ACL, 2023:1236-1270.
[7] ZHOU J, DENG J, MI F, et al. Towards identifying social bias in dialog systems:frame, datasets, and benchmarks[C] //Findings of the association for computational linguistics:2022 conference on empirical methods in natural language Processing. Abu Dhabi:ACL, 2022:3576-3591.
[8] DENG J, ZHOU J, SUN H, et al. COLD:a benchmark for Chinese offensive language detection[C] //Proceedings of the2022 conference on empirical methods in natural language processing. Abu Dhabi:ACL, 2022:11580-11599.
[9] JIANG A, YANG X, LIU Y, et al. SWSR:a Chinese dataset and lexicon for online sexism detection[J]. Online social networks and media, 2022, 27:100182.
[10] WANG H, LI M, LU J, et al. CCPC:A hierarchical Chinese corpus for patronizing and condescending language detection[C] //Proceedings of CCF international conference on natural language processing and Chinese computing. Foshan:CCF, 2023:640-652.
[11] ZHAO J, FANG M, SHI Z, et al. CHBias:bias evaluation and mitigation of Chinese conversational language models[C] //Proceedings of the 61st annual meeting of the association for computational linguistics. Toronto:ACL, 2023:13538-13556.
[12] XI Z, ZHENG R, GUI T. Safety and ethical concerns of large language models[C] //Proceedings of the 22nd Chinese national conference on computational linguistics. Harbin:Chinese information processing society of China, 2023:9-16.
[13] HUANG Y, XIONG D. CBBQ:a Chinese bias benchmark dataset curated with human-AI collaboration for large language models[C] //Proceedings of the 2024 joint international conference on computational linguistics, language resources and evaluation. Turin:LREC, 2024:2917-2929.
[14] NAOUS T, RYAN M J, RITTER A, et al. Having beer after prayer? Measuring cultural bias in large language models[C] //Proceedings of the 62nd annual meeting of the association for computational linguistics(volume 1:Long Papers). Bangkok:ACL, 2024:16366-16393.
[15] RAJ C, MUKHERJEE A, CALISKAN A, et al. Breaking bias,building bridges:Evaluation and mitigation of social biases in LLMs via contact hypothesis[C] //Proceedings of the AAAI/ACM Conference on AI, ethics, and society. San Jose:ACM, 2024:1180-1189.
[16] SAHOO N, KULKARNI P, AHMAD A, et al. IndiBias:a benchmark dataset to measure social biases in language models for Indian context[C] //Proceedings of the 2024 Conference of the north American chapter of the association for computational linguistics:Human language technologies(volume 1:Long Papers). Mexico City:ACL, 2024:8778-8798.
[17] 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. Stroudsburg:ACL,2020:1953-1967.
[18] 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.
[19] PARRISH A, CHEN A, NANGIA N, et al. BBQ:a hand-built bias benchmark for question answering[C] //Findings of the 60th annual meeting of the association for computational linguistics.Dublin:ACL, 2022:2086-2105.
[20] AHMAD W, ZHANG Z, MA X, et al. On difficulties of cross-lingual transfer with order differences:a case study on dependency parsing[C] //Proceedings of the 2019 conference of the north American chapter of the association for computational linguistics:Human language technologies, volume 1(Long and Short Papers). Minneapolis:ACL, 2019:2440-2452.
[21] ZHAO J, MUKHERJEE S, HOSSEINI S, et al. Gender bias in multilingual embeddings and cross-lingual transfer[C] //Proceedings of the 58th annual meeting of the association for computational linguistics. Stroudsburg:ACL, 2020:2896-2907.
[22] 刘文炎,沈楚云,王祥丰,等.可信机器学习的公平性综述[J].软件学报, 2021, 32(5):1404-1426.(LIU W Y, SHEN C Y, WANG X F, et al. Survey on fairness in trustworthy machine learning[J]. Journal of software, 2021, 32(5):1404-1426.)
[23] BOURDIEU P. Masculine domination[M]. Stanford:Stanford University Press, 2001.
[24] FEDERICI S. Caliban and the witch:Women, the body and primitive accumulation[M]. New York:Autonomedia, 2004.
[25] MYERS D. Social psychology[M]. St. Petersburg:Piter, 2003.
[26] 贾凤芹.大学生性别偏见内隐与外显双系统研究[D].苏州:苏州大学, 2013.(JIA F Q. A research on implicit and explicit gender prejudice of college students[D]. Suzhou:Soochow University, 2013.)
[27] 孟令宇.从算法偏见到算法歧视:算法歧视的责任问题探究[J].东北大学学报(社会科学版), 2022, 24(1):1-9.(MENG L Y. From algorithm bias to algorithm discrimination:research on the responsibility of algorithmic discrimination[J]. Journal of Northeastern University(social science), 2022, 24(1):1-9.)
[28] 谢永江,杨永兴.人工智能时代下的算法歧视及其治理研究[J].北京邮电大学学报(社会科学版), 2022, 24(5):18-25.(XIE Y J, YANG Y X. Algorithmic discrimination and its governance in era of artificial intelligence[J]. Journal of Beijing University of Posts and Telecommunications(social sciences edition), 2022,24(5):18-25.)
[29] HOVY D, PRABHUMOYE S. Five sources of bias in natural language processing[J]. Language and linguistics compass, 2021,15(8):e12432.
[30] RUDINGE R, MAY C, VAN DURME B. Social bias in elicited natural language inferences[C] //Proceedings of the 1st association for computational linguistics workshop on ethics in natural language processing. Valencia:ACL, 2017:74-79.
[31] SINGH V K, CHAYKO M, INAMDAR R, et al. Female librarians and male computer programmers? gender bias in occupational images on digital media platforms[J]. Journal of the Association for Information Science and Technology, 2020,71(11):1281-1294.
[32] GEVA M, GOLDBER Y, BERANT J. Are we modeling the task or the annotator? An investigation of annotator bias in natural language understanding datasets[C] //Proceedings of the 2019conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing. Hong Kong:ACL, 2019:1161-1166.
[33] MEMON S A, DHAMYAL H, WRIGHT O, et al. Detecting gender differences in perception of emotion in crowdsourced data[J]. ArXiv preprint, 2019, arXiv:1910.11386.
[34] DURME B D V. Extracting implicit knowledge from text[D].Rochester:University of Rochester, 2009.
[35] 周燚儒.人工智能时代算法歧视与平等权保护研究[D].济南:山东大学, 2020.(ZHOU Y R. Research on algorithm discrimination and equal right protection in the era of artificial intelligence[D]. Ji’nan:Shandong University, 2020.)
[36] DE-ARTEAGA M, ROMANOV A, WALLACH H, et al. Bias in bios:a case study of semantic representation bias in a highstakes setting[C] //Proceedings of the 2019 Association for Computing Machinery Conference on Fairness, Accountability,and Transparency. Atlanta:ACM, 2019:120-128.
[37] KIRITCHENKO S, MOHAMMAD S M. Examining gender and race bias in two hundred sentiment analysis systems[C] //Proceedings of the 7th joint conference on lexical and computational semantics. New Orleans:ACM, 2018:43-53.
[38] THELWALL M. Gender bias in sentiment analysis[J]. Online information review, 2018, 42(1):45-57.
[39] BHASKARAN J, BHALLAMUDI I. Good secretaries, bad truck drivers? Occupational gender stereotypes in sentiment analysis[C] //Proceedings of the 1st workshop on gender bias in natural language processing. Florence:ACL, 2019:62-68.
[40] SHENG E, CHANG K W, NATARAJAN P, et al. The woman worked as a babysitter:on biases in language generation[C] //Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing. Hong Kong:ACL, 3407-3412.
[41] GEHMANS, GURURANGANS, SAPM, et al.RealToxicityPrompts:evaluating neural toxic degeneration in language models[C] //Findings of the association for computational linguistics:2020 conference on empirical methods in natural language processing. Stroudsburg:ACL, 2020:3356-3369.
[42] KIRK H R, JUN Y, IQBAL H, et al. Bias out-of-the-box:an empirical analysis of intersectional occupational biases in popular generative language models[C] //Proceedings of advances in neural information processing systems. New York:Curran Associates, 2021:2611-2624.
[43] VENKIT P N, GAUTAM S, PANCHANADIKAR R, et al.Nationality bias in text generation[C] //Proceedings of the17th conference of the European chapter of the association for computational linguistics. Dubrovnik:ACL, 2023:116-122.
[44] SUTSKEVER I, VINYALS O, LE Q V. Sequence to sequence learning with neural networks[C] //Proceedings of advances in neural information processing systems. Montreal:Curran Associates, 2014:3104-3112.
[45] DEVLIN J, CHANG M W, LEE K, et al. Bert:Pre-training of deep bidirectional transformers for language understanding[C] //Proceedings of the 2019 conference of the north American chapter of the association for computational linguistics:human language technologies, Volume 1(long and short papers).Minneapolis:ACL, 2019:4171-4186.
[46] SUN Y, WANG S, LI Y, et al. Ernie:enhanced representation through knowledge integration[J]. ArXiv preprint, 2019, arXiv:1904.09223.
[47] BROWN T, MANN B, RYDER N, et al. Language models are few-shot learners[C] //Proceedings of advances in neural information processing systems. Vancouver:Curran Asscoiates,2020:1877-1901.
[48] VASWANI A, SHAZEER N, PARMAR N, et al. Attention is all you need[C] //Proceedings of advances in neural information processing systems. Long Beach:Curran Associates, 2017:6000-6010.
[49] OUYANG L, WU J, JIANG X, et al. Training language models to follow instructions with human feedback[C] //Proceedings of advances in neural information processing systems. New Orleans:Curran Associates, 2022:27730-27744.
[50] KAMRUZZAMAN M, SHOVON M M I, KIM G L.Investigating subtler biases in LLMs:ageism, beauty,institutional, and nationality bias in generative models[J]. ArXiv preprint, 2023, ArXiv:2309.08902.
[51] CHENG M, DURMUS E, JURAFSKY, et al. 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. Toronto:ACL,2023:1504-1532.
[52] GUPTA S, SHRIVASTAVA V, DESHPANDE A, et al. Bias runs deep:implicit reasoning biases in persona-assigned LLMs[C] //Proceedings of the twelfth international conference on learning representations. Vienna:ICLR, 2024.
[53] ZHAO J, WANG T, YATSKAR M, et al. Gender bias in coreference resolution:evaluation and debiasing methods[C] //Proceedings of the 2018 conference of the north American chapter of the association for computational linguistics:human language technologies. New Orleans:ACL, 2018:2740-2750.
[54] ZHU S C, LIU Y. Offensiveness analysis of Chinese group addressing terms and dataset construction[C] //Workshop on Chinese lexical semantics. Singapore:Springer Nature Singapore,2023:342-356.
Options
Outlines

/

〈 〉

    Copyrights © LIS Press Co., Ltd. Reproduction is prohibited without permission.