Anthropic’s Persona dataset

该数据集包含了多种不同类型的人格特征，总计有99种不同的人格。每个人格包括500条与该人格特征一致的陈述和500条与之相反的陈述。数据集被划分为700个训练样本和300个测试样本，同时保持了标签比例。关于数据集的更多详细信息，请参阅论文附录A。规模上，每种人格包含1000条陈述（其中500条支持人格特征，500条反对）。该任务旨在进行人格引出及行为偏好评估。

This dataset contains various types of personality traits, with a total of 99 distinct personality types. For each personality type, there are 500 statements consistent with the corresponding personality trait and 500 opposing statements. The dataset is split into 700 training samples and 300 test samples while preserving the label distribution. For more detailed information about this dataset, please refer to Appendix A of the paper. In terms of scale, each personality type includes 1000 statements in total (500 supporting the trait and 500 opposing it). This task aims at personality elicitation and behavioral preference evaluation.