SBIC

该数据集名为SBIC，其中训练集包含了34,000份文档，这些文档被归类在人们对他人的社会偏见和刻板印象的类别下。任务包括识别冒犯性、意图、下流、群体、毒性和性暗示内容。规模上，该数据集共有34,000份文档，任务类型为多标签分类，包含二元任务。

This dataset is designated as SBIC. Its training set includes 34,000 documents categorized under the context of people's social biases and stereotypes toward others. The associated tasks involve detecting offensive content, intent, profanity, targeted groups, toxic language, and sexually suggestive materials. Overall, the full dataset contains 34,000 documents, with the task paradigm being multi-label classification that also encompasses binary classification tasks.

搜集汇总

背景与挑战

背景概述

SBIC（Social Bias Inference Corpus）是一个包含150k条社交媒体帖子结构化标注的数据集，专注于捕捉语言中隐含的社会偏见，覆盖超过34k条关于一千个人口群体的偏见暗示。该数据集旨在通过Social Bias Frames方法，更全面地分析和解释语言中的冒犯性、说话者意图及社会动态，以支持自动检测和AI辅助应用，同时强调伦理考量。