JPHARMABENCH
收藏资源简介:
JPHARMABENCH是一个为评估日本医药领域的语言模型而构建的基准数据集,包括三个子数据集:YakugakuQA、NayoseQA和SogoCheck。YakugakuQA基于日本国家药师资格考试,包含3,021个问题;NayoseQA用于评估药物名称和活性成分的标准化,包含34,769个问题;SogoCheck用于检测配对文本之间的逻辑或事实不一致,包含200个问题。这些数据集基于公开可用的数据构建,旨在反映医药公司实际操作中遇到的各种情况,如监管和文书工作。这些数据集为日本医药领域的语言模型评估提供了宝贵的资源。
JPHARMABENCH is a benchmark dataset constructed for evaluating language models in the Japanese pharmaceutical domain, encompassing three sub-datasets: YakugakuQA, NayoseQA, and SogoCheck. YakugakuQA is derived from Japan's National Pharmacist Qualification Examination and contains 3,021 questions. NayoseQA is designed to evaluate the standardization of drug names and active ingredients, comprising 34,769 questions. SogoCheck is utilized to detect logical or factual inconsistencies between paired texts, with 200 questions in total. All these datasets are built upon publicly available data, aiming to reflect various scenarios encountered in actual pharmaceutical company operations, such as regulatory work and documentation tasks. These datasets provide a valuable resource for language model evaluation in the Japanese pharmaceutical field.

- 1A Japanese Language Model and Three New Evaluation Benchmarks for Pharmaceutical NLPEQUES Inc., The University of Tokyo, University of Tsukuba · 2025年



