PANDA – Paired Anti-hate Narratives Dataset from Asia
收藏资源简介:
PANDA数据集是首个专注于中文反仇恨言论的语料库,由佛罗里达大学和Lingua的研究团队创建。该数据集包含2974条仇恨言论与反仇恨言论的配对数据,旨在解决中文社交媒体中仇恨言论的对抗问题。数据来源包括多个开源中文仇恨言论数据集,如COLD、SWSR和CHSD。通过使用LLM-as-a-Judge和模拟退火算法生成反仇恨言论,并经过人工验证,确保数据的质量和上下文相关性。该数据集为中文反仇恨言论的生成和评估提供了重要资源,适用于自然语言处理领域的研究和应用。
PANDA Dataset is the first corpus focused on Chinese anti-hate speech, created by research teams from the University of Florida and Lingua. This dataset contains 2,974 paired instances of hate speech and anti-hate speech, aiming to address the adversarial problem of hate speech on Chinese social media. Its data sources cover multiple open-source Chinese hate speech datasets, including COLD, SWSR and CHSD. Anti-hate speech were generated using the LLM-as-a-Judge framework and simulated annealing algorithm, and underwent manual verification to ensure data quality and contextual relevance. This dataset provides an important resource for the generation and evaluation of Chinese anti-hate speech, applicable to both research and practical applications in the field of natural language processing.

- 1PANDA -- Paired Anti-hate Narratives Dataset from Asia: Using an LLM-as-a-Judge to Create the First Chinese Counterspeech Dataset佛罗里达大学, Lingua · 2025年



