Biasly

Biasly数据集是由Mila - 魁北克人工智能研究所与多学科专家合作创建，专注于捕捉电影字幕中微妙的性别歧视表达。该数据集包含10,000条经过专家标注的电影字幕，涵盖了北美电影中的日常语言中的性别歧视。创建过程中，专家们精心筛选和培训了标注者，确保数据集的高质量和社交责任感。Biasly数据集适用于多种NLP任务，如分类、严重性评分回归和文本重写，旨在帮助检测和减轻微妙的性别歧视，推动AI在社会公益方面的应用。

The Biasly Dataset was developed in collaboration with multidisciplinary experts by Mila – Quebec Artificial Intelligence Institute, focusing on capturing subtle sexist expressions within movie subtitles. It consists of 10,000 expert-annotated movie subtitles covering sexist language in daily discourse from North American films. During the creation process, experts meticulously selected and trained annotators to guarantee the dataset's high quality and social responsibility. The Biasly Dataset supports multiple NLP tasks including classification, severity scoring regression, and text rewriting, with the goal of assisting in the detection and mitigation of subtle sexism and advancing the application of AI in social good endeavors.