KoGEM
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/SungHo3268/KoGEM
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为KoGEM，包含了1500多道多选题，覆盖了五大类别和16个子类别，旨在评估大型语言模型和人类在韩语语言能力方面的表现。该数据集基于韩国官方的职能能力考试，并包含了用于评估的众包人类表现数据。规模上，数据集包含了1500组问答对，其任务是对语言能力进行评估。
The dataset named KoGEM includes over 1,500 multiple-choice questions spanning 5 primary categories and 16 subcategories. It is designed to assess the performance of both Large Language Models (LLMs) and human participants in Korean language proficiency. Built upon South Korea's official vocational competency examination, this dataset also incorporates crowdsourced human performance data for evaluation purposes. With a total of 1,500 question-answer pairs, its core task is to evaluate language proficiency.
提供机构：
Authors of the paper
