MSMGenderBias
收藏资源简介:
MSMGenderBias是一个公开可用的数据集,包含MS MARCO Passage Ranking集合中一部分文档的性别偏见标注。该数据集旨在促进对性别偏见检测和测量的研究,特别是利用大型语言模型(LLMs)来检测和衡量性别偏见。数据集的创建基于LLM驱动的性别偏见检测方法和新的性别公平度度量标准Class-wise Weighted Exposure(CWEx)。通过整合LLM驱动的偏见检测、改进的公平度度量标准和性别偏见标注,该数据集提供了一个更强大的框架,用于分析和减轻信息检索系统中的偏见。
MSMGenderBias is a publicly available dataset containing gender bias annotations for a subset of documents from the MS MARCO Passage Ranking collection. This dataset aims to advance research on gender bias detection and measurement, particularly leveraging Large Language Models (LLMs) for such tasks. It was developed based on an LLM-driven gender bias detection approach and a novel gender fairness metric, Class-wise Weighted Exposure (CWEx). By integrating LLM-driven bias detection, improved fairness metrics, and gender bias annotations, this dataset provides a more robust framework for analyzing and mitigating bias in information retrieval systems.
数据集概述:MSMGenderBias
📌 基本信息
- 数据集名称:MSMGenderBias
- 相关论文:Towards Fair Rankings: Leveraging LLMs for Gender Bias Detection and Measurement
- 许可证:CC BY 4.0
- 联系方式:maryamalsadat.mousavian@usi.ch
📊 数据集内容
- 文件名称:MSMGenderBias.csv
- 数据来源:MS MARCO Passage Ranking collection(子集)
- 字段说明:
doc_id:MS MARCO passage identifiergender_bias:性别偏见标签- N — 中性(无偏见)
- M — 男性偏见
- F — 女性偏见
⚠️ 注意事项
- 原始MS MARCO数据集不包含在此仓库中,需单独从MS MARCO官网申请访问权限。
📚 引用格式
bibtex @inproceedings{mousavian2025towards, title={Towards Fair Rankings: Leveraging LLMs for Gender Bias Detection and Measurement}, author={Maryam Mousavian, Zahra Abbasiantaeb, Mohammad Aliannejadi, Fabio Crestani}, booktitle={ICTIR}, year={2025} }
🙏 致谢
- 感谢微软MS MARCO团队为研究社区提供数据集。

- 1Towards Fair Rankings: Leveraging LLMs for Gender Bias Detection and MeasurementUniversità della Svizzera italiana & University of Amsterdam · 2025年



