KISTI-KONI/ScholarBench

ScholarBench是一个双语的学术推理能力评估基准，包含韩语和英语两种语言。它涵盖八个研究领域，包括商业研究、化学生物科学、工程、物理与数学、地球与生命科学、医学科学、社会专业研究和自由艺术与社会科学。数据集包含五种问题类型：总结、简答题、选择题、多选题和判断题。该数据集旨在评估大型语言模型在特定学术领域内的抽象、理解和逻辑推理能力。

ScholarBench is a bilingual benchmark for evaluating the academic reasoning capabilities of large language models in domain-specific contexts, including both Korean and English languages. It covers eight research domains: Business Studies, Chemical Biosciences, Engineering, Physics & Mathematics, Earth & Life Sciences, Medical Science, Socio-Professional Studies, and Liberal Arts & Social Sciences. The dataset includes five types of tasks: summarization, short answer, multiple choice, multiple selection, and true/false. It is designed to assess the abilities of LLMs in abstraction, comprehension, and logical inference within specific academic fields.