TechWolf/Skill-extraction-Tech-graded
收藏资源简介:
这是一个基于TechWolf/skill-extraction-tech数据集的句子,针对ESCO v1.1.0技能分类体系进行分级相关性标注的数据集。数据集遵循BEIR格式,包含三个部分:queries(查询句子,共75条,包含句子ID和文本)、corpus(ESCO技能条目,共13,891条,包含技能URI、英文标签、英文描述和版本信息)和qrels(查询-技能相关性评分,共1,041,825条,包含查询ID、技能ID和评分)。评分采用0-4级标准:0表示完全不相关,1表示技能领域正确但未在句子中提及,2表示可推荐但非核心,3表示强相关但隐含,4表示明确展示或请求。验证集提供完整分级标注,测试集目前为二元相关性标注(0或1),未来将发布完整分级标注。数据集用于检索和技能提取任务,语言为英语,基于CC BY 4.0许可证。
Graded-relevance annotations for sentences from TechWolf/skill-extraction-tech against the ESCO v1.1.0 skill taxonomy. The dataset follows the BEIR convention and includes three configs: queries (75 sentences with IDs and text), corpus (13,891 ESCO skill entries with URIs, English labels, descriptions, and version), and qrels (1,041,825 query-skill relevance scores). Scores range from 0-4: 0 for totally unrelated, 1 for correct domain but not mentioned, 2 for plausible but not core, 3 for strongly relevant but implied, and 4 for explicitly demonstrated. The validation split has full graded annotations, while the test split currently has binary relevance (0 or 1), with full graded annotations to be released later. It is used for retrieval and skill extraction tasks, in English, under CC BY 4.0 license.




