EusExams-v2
收藏资源简介:
EusExams-v2 是一个用于评估和训练巴斯克语(eu)和西班牙语(es)语言模型的问答数据集,专门针对巴斯克地区公共部门的考试场景。该数据集收集了来自多个巴斯克公共机构(包括公共卫生系统 Osakidetza、巴斯克政府、毕尔巴鄂和加斯泰斯市议会以及巴斯克大学 UPV/EHU)的公共服务考试题目,旨在为相关职位(如行政和助理)的备考提供资源。数据集内容为多项选择题,每个问题包含2到4个选项和一个正确答案。V2 版本相较于原始数据集进行了重要改进:进行了去重处理以消除同一类别内不同考试中的重复问题;移除了无效答案的问题以确保评估的鲁棒性;将众多细粒度测试文件重组为四个高级领域(议会、政府、健康、大学);为每个问题添加了 `ids` 和 `links` 字段以追踪其原始考试来源;并新增了2026年的西班牙语健康考试题目。数据集总规模为 27,442 个唯一问题,按语言和领域分布如下:西班牙语部分共 16,304 题(议会 928,政府 2,530,健康 11,387,大学 1,459),巴斯克语部分共 11,138 题(议会 895,政府 2,531,健康 6,351,大学 1,361)。数据格式为 JSONL,每个样本包含问题文本(question)、候选答案列表(candidates)、正确答案索引(answer,从0开始)、原始问题ID列表(ids)和来源链接列表(links)。该数据集适用于问答、多项选择等自然语言处理任务,尤其侧重于法律和公共管理领域的语言理解评估。
EusExams-v2 is a question answering dataset designed for evaluating and training language models in Basque (eu) and Spanish (es), specifically targeted at public sector exam scenarios in the Basque Country. This dataset collects public service exam questions from multiple Basque public institutions, including the public health system Osakidetza, the Basque Government, the city councils of Bilbao and Gasteiz, and the University of the Basque Country UPV/EHU, aiming to provide resources for exam preparation of relevant positions such as administrative and assistant roles. The dataset comprises multiple-choice questions, each with 2 to 4 options and a correct answer. Compared with the original dataset, Version 2 has undergone substantial improvements: duplicate questions across different exams within the same category are removed via deduplication; questions with invalid answers are eliminated to ensure robust evaluation; numerous fine-grained test files are reorganized into four high-level domains, namely Parliament, Government, Health, and University; `ids` and `links` fields are added to each question to trace its original exam source; and 2026 Spanish health exam questions are additionally included. The total scale of the dataset is 27,442 unique questions, distributed by language and domain as follows: the Spanish subset contains 16,304 questions (Parliament: 928, Government: 2,530, Health: 11,387, University: 1,459), while the Basque subset includes 11,138 questions (Parliament: 895, Government: 2,531, Health: 6,351, University: 1,361). The data is stored in JSONL format, with each sample containing the question text (`question`), list of candidate answers (`candidates`), correct answer index (`answer`, 0-indexed), list of original question IDs (`ids`), and list of source links (`links`). This dataset is applicable to natural language processing tasks such as question answering and multiple choice, with a particular focus on language understanding evaluation in the legal and public administration domains.
数据集概述:EusExams-v2
EusExams-v2 是一个用于巴斯克公共部门考试准备的问答题数据集,包含多项选择题,每个问题有2到4个选项和一个正确答案。该数据集由HiTZ研究中心和IXA研究组(巴斯克大学UPV/EHU)整理。
主要特性
- 语言:巴斯克语 (eu) 和西班牙语 (es)
- 任务:问答 (question-answering) 和多项选择 (multiple-choice)
- 领域:法律 (legal)
- 许可协议:cc-by-sa-4.0
- 数据规模:10,000 < n < 100,000
V2版本改进
- 去重:移除了同一类别内考试中相同的题目。
- 空值过滤:删除了缺少有效答案的问题。
- 分组:将众多考试文件重组为4个高级领域:
council(市政厅)、government(政府)、health(健康)和university(大学)。 - 来源追踪:为每个问题添加了
ids和links字段,以追溯原始考试。 - 新增数据:新增了西班牙语的2026年健康考试数据。
数据集统计
总共有 27,442 个不重复的问题,按语言和域分布如下:
| 域 (Domain) | 西班牙语 (es) | 巴斯克语 (eu) |
|---|---|---|
| Council (市政厅) | 928 | 895 |
| Government (政府) | 2,530 | 2,531 |
| Health (健康) | 11,387 | 6,351 |
| University (大学) | 1,459 | 1,361 |
| 总计 | 16,304 | 11,138 |
数据配置
数据集包含8个子配置,每个配置仅包含 test 拆分,对应语言与域的组合:
es_counciles_governmentes_healthes_universityeu_councileu_governmenteu_healtheu_university
数据示例
每个问题以JSON格式呈现,包含问题文本、候选答案列表、正确答案索引、原始考试的ID列表以及相关链接。例如,一个巴斯克语问题会有一个对应的英语翻译。
相关资源




