GenderPair

该数据集是一个创新的对偶基准测试，旨在全面评估大型语言模型中的性别偏见问题。它提供了标准化和现实化的评估方法，涵盖了包括跨性别和非二元性别个体在内的、之前被忽视的性别群体。GenderPair基准测试涵盖了所有提议的性别平等基准标准维度，包括包容性、多样性、可解释性、客观性、鲁棒性和现实性，旨在对大型语言模型中的性别偏见进行评估和减少。

This dataset is an innovative pairwise benchmark designed to comprehensively evaluate gender bias in large language models. It provides standardized and realistic assessment methodologies, covering previously overlooked gender groups including transgender and non-binary individuals. The GenderPair Benchmark encompasses all proposed gender equality benchmark dimensions, including inclusivity, diversity, interpretability, objectivity, robustness, and realism, aiming to evaluate and mitigate gender bias in large language models.