CultureBank

该数据集名为CultureBank，它是一个由TikTok和Reddit上的用户自我叙述构建的知识库，其中包含了关于文化描述的多元观点以及具体文化情境。该数据集涵盖了730个文化群体和36个话题，且经人类标注者评估，文化描述的质量获得了0.8的Kappa评分。数据集的规模包括34,000个帖子和720,000条英文评论；此外，还有12,000个来自TikTok的文化描述。这项数据的任务旨在评估语言模型的文化意识，并对语言模型在文化任务上进行微调。This dataset, named CultureBank, is a knowledge base constructed from user self-narratives collected from TikTok and Reddit, encompassing diverse perspectives on cultural descriptions and specific cultural contexts. It covers 730 cultural groups and 36 topics, and the quality of its cultural descriptions achieved a Cohen's Kappa score of 0.8 as evaluated by human annotators. The dataset includes 34,000 posts and 720,000 English comments; additionally, there are 12,000 cultural descriptions sourced from TikTok. The tasks based on this dataset aim to assess the cultural awareness of language models and facilitate fine-tuning of language models on cultural tasks.

搜集汇总

背景与挑战

背景概述

CultureBank数据集是一个专注于文化行为和社会规范的多维度文本数据集，包含约23k行数据，涵盖美国文化中的饮食、礼仪、家庭、沟通等多个主题。数据集通过详细的字段描述文化场景、行为、关系及评估问题，适用于文化理解、自然语言处理等研究和应用。