Crisp-Unimib/JobSet
收藏资源简介：
--- license: cc-by-4.0 task_categories: - text-classification - zero-shot-classification language: - en pretty_name: JobSet size_categories: - 10K<n<100K --- # Dataset Card for JobSet <!-- Provide a quick summary of the dataset. --> **JobSet** is a novel, synthetic dataset of Online Job Advertisements designed to enhance machine learning applications in labour market intelligence. ## Dataset Details ### Dataset Description <!-- Provide a longer summary of what this dataset is. --> The use of online services for advertising job positions has grown in the last decade, thanks to the ability of Online Job Advertisements (OJAs) to observe the labour market in near real-time, predict new occupation trends, identify relevant skills, and support policy and decision-making activities. Unsurprisingly, 2023 was declared the Year of Skills by the EU, as skill mismatch is a key challenge for European economies. In such a scenario, machine learning-based approaches have played a key role in classifying job ads and extracting skills according to well-established taxonomies. However, the effectiveness of ML depends on access to annotated job advertisement datasets, which are often limited and require time-consuming manual annotation. The lack of OJA annotated benchmarks representative of the real online OJA and skills distributions is currently limiting advances in skill intelligence. To deal with this, we propose **JobGen**, which leverages Large Language Models (LLMs) to generate synthetic OJAs. We use real OJAs collected from an EU project and the ESCO taxonomy to represent job market distributions accurately. JobGen enhances data diversity and semantic alignment, addressing common issues in synthetic data generation. The resulting dataset, **JobSet**, provides a valuable resource for tasks like skill extraction and job matching and is openly available to the community. - **Curated by:** CRISP research centre https://crispresearch.it/ - **Language(s) (NLP):** English - **License:** Creative Commons Attribution 4.0 International (CC BY 4.0) ### Dataset Sources <!-- Provide the basic links for the dataset. --> - **Repository:** https://github.com/Crisp-Unimib/JobGen - **Zenodo:** https://doi.org/10.5281/zenodo.11454052 - **Paper [optional]:** [ACCEPTED AT SAC 25] ## Uses <!-- Address questions around how the dataset is intended to be used. --> ### Direct Use <!-- This section describes suitable use cases for the dataset. --> The dataset has been used for tasks related to labour market analysis, specifically for training and evaluating machine learning models in skill extraction and job matching. These tasks leverage the synthetic job advertisements to improve the accuracy and performance of models designed to analyse job market trends, identify required skills, and facilitate better job seeker-opportunity matches. The dataset could be used for tasks beyond its initial scope, including skill gap analysis, job market segmentation, and job recommendation systems. ### Out-of-Scope Use <!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. --> While the JobSet dataset offers numerous advantages for various labour market analytics tasks, there are certain applications for which it is unsuitable. Future users should avoid using the dataset in automated hiring systems, which could lead to unfair treatment or biased outcomes. The synthetic nature of the data and potential biases in its generation process makes it unsuitable for critical employment decisions without significant validation and bias correction. ## Dataset Structure <!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. --> Each instance in the dataset represents a synthetic Online Job Advertisement consisting of synthetic text and the corresponding ESCO occupation and skills. The dataset is designed to mimic real-world job advertisements, capturing the diversity and specifics of different job roles and their associated skills. Every instance is an individual job advertisement. Starting with 396 ESCO IV digit-level occupations, we generated 15,469 unique job postings reflecting real-world online frequency distributions. Our balancing approach achieved this, ensuring each occupation is adequately represented with an average of 39 generations per occupation class and ten average skills per generation, totalling 8,374 unique skills. JobSet comprises 4.2 million tokens across all job ads, averaging 214.8 words and 264 tokens per job advertisement. ## Dataset Creation #### Data Collection and Processing <!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. --> Our framework JobGen consists of four primary components: (1) occupation-skill combination generation using the ESCO taxonomy (European skills, competences, qualifications, and occupations) to represent skills and occupations and the WIH dataset to reflect real-world data distributions, (2) content generation through an LLM based on the combinations above, and (3) fitness evaluation to ensure diversity and semantic consistency. Lastly, the process is iterated until convergence if the diversity is unsatisfactory (4). Please refer to our research paper for more information about the full process. #### Personal and Sensitive Information <!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. --> The dataset does not contain data that might be considered confidential since it comprises entirely synthetic job advertisements generated using LLMs and does not include real data from individuals or any non-public communications. Additionally, all data used in the generation process from the ESCO taxonomy and any supplemental sources are anonymised and do not contain personally identifiable information or confidential content. The dataset does not contain data that might be viewed as offensive, insulting, threatening, or anxiety-causing. Since the dataset comprises entirely synthetic data generated by computational methods, which does not involve human subjects or personal data, it did not necessitate an ethical review process. ## Citation [optional] <!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. --> **BibTeX:** [More Information Needed] **APA:** Colombo, S., D'Amico, S., Malandri, L., Mercorio, F., & Seveso, A. (2024). JobSet (1.0.1) [Data set]. Zenodo. https://doi.org/10.5281/zenodo.11484885 ## Dataset Card Contact Andrea Seveso - andrea.seveso@unimib.it Antonio Serino - antonio.serino@unimib.it
--- 许可证：CC BY 4.0 任务类别： - 文本分类（text-classification） - 零样本分类（zero-shot-classification） 语言： - 英语（en） 数据集名称：JobSet 数据规模类别： - 10K<n<100K --- # JobSet 数据集卡片 <!-- 请提供数据集的简要概述。 --> **JobSet** 是一款全新的合成式在线招聘广告数据集，旨在助力劳动力市场智能领域的机器学习应用发展。 ## 数据集详情 ### 数据集概述 <!-- 请提供关于数据集的更详细说明。 --> 近十年来，借助在线招聘广告（Online Job Advertisements, OJAs）能够近乎实时地观测劳动力市场、预测新兴职业趋势、识别关键技能并支撑政策制定与决策活动的特性，在线发布职位招聘的服务应用大幅增长。 2023年欧盟将其定为“技能之年”，这并不意外，因为技能错配是欧洲经济体面临的核心挑战之一。在此背景下，基于机器学习的方法在按既定分类体系对招聘广告进行分类、提取技能信息方面发挥了关键作用。然而，机器学习模型的有效性依赖于带标注的招聘广告数据集，但这类数据集往往数量有限，且需要耗费大量人力进行手动标注。 当前，缺乏能够反映真实在线招聘广告与技能分布的带标注基准数据集，这限制了技能智能领域的研究进展。 为解决这一问题，我们提出了**JobGen**框架，该框架借助大语言模型（Large Language Model, LLM）生成合成式在线招聘广告。我们使用从欧盟项目中收集的真实在线招聘广告，以及ESCO分类体系来精准还原劳动力市场的分布情况。 JobGen提升了数据的多样性与语义对齐度，解决了合成数据生成过程中的常见问题。最终得到的数据集**JobSet**为技能提取、职位匹配等任务提供了宝贵的资源，并对社区公开可用。 - **整理方**：CRISP研究中心 https://crispresearch.it/ - **自然语言处理所用语言**：英语 - **许可证**：知识共享署名4.0国际许可协议（CC BY 4.0） ### 数据集来源 <!-- 请提供数据集的基础链接。 --> - **代码仓库**：https://github.com/Crisp-Unimib/JobGen - **Zenodo存档**：https://doi.org/10.5281/zenodo.11454052 - **相关论文**：[已被SAC 25收录] ## 数据集用途 ### 直接用途 <!-- 本节描述数据集适用的用例场景。 --> 本数据集已被用于劳动力市场分析相关任务，具体而言，用于训练和评估技能提取、职位匹配领域的机器学习模型。 这些任务依托合成式招聘广告，旨在提升用于分析劳动力市场趋势、识别所需技能、优化求职者与岗位机会匹配度的模型的准确率与性能。 本数据集还可用于超出初始设计范围的任务，包括技能缺口分析、劳动力市场细分以及职位推荐系统等。 ### 超出适用范围的用途 <!-- 本节说明数据集的误用、恶意使用以及不适合的应用场景。 --> 尽管JobSet数据集为各类劳动力市场分析任务提供了诸多优势，但仍有部分应用场景并不适配。未来使用者应避免将该数据集用于自动化招聘系统，此类应用可能导致不公平待遇或产生偏差结果。由于数据为合成生成，且生成过程中可能存在偏差，若未经过充分的验证与偏差校正，该数据集不适用于关键的雇佣决策场景。 ## 数据集结构 <!-- 本节将介绍数据集的字段信息，以及数据集划分标准、数据点间关联关系等额外结构细节。 --> 数据集中的每个实例均代表一条合成式在线招聘广告，包含合成文本以及对应的ESCO职业与技能信息。本数据集旨在模拟真实的招聘广告，覆盖不同职位的多样性与具体特征，以及其关联的技能信息。每条实例均为独立的招聘广告。 我们以396个ESCO四级职业分类为基础，生成了15469条独特的招聘信息，还原了真实的在线岗位发布频率分布。我们通过均衡化处理实现了这一目标，确保每个职业分类都得到充分覆盖：每个职业分类平均生成39条招聘信息，每条招聘信息平均关联10项技能，总计涵盖8374项独特技能。 JobSet所有招聘广告总计包含420万Token，单条招聘广告平均包含214.8个单词与264个Token。 ## 数据集构建 #### 数据收集与处理流程 <!-- 本节将描述数据收集与处理流程，例如数据选择标准、过滤与归一化方法、所用工具与库等。 --> 我们的JobGen框架包含四个核心组件：(1) 基于欧洲技能、能力、资格与职业分类体系（ESCO）生成职业-技能组合，并依托WIH数据集还原真实数据分布；(2) 基于上述组合通过大语言模型（LLM）生成招聘广告内容；(3) 适配性评估，确保生成内容的多样性与语义一致性；(4) 若多样性未达要求，则迭代上述流程直至收敛。 如需了解完整流程的更多细节，请参阅我们的研究论文。 #### 个人与敏感信息说明 <!-- 请说明数据集是否包含可能被视为个人、敏感或隐私的数据（例如泄露地址、唯一可识别的姓名或别名、种族或族裔出身、性取向、宗教信仰、政治观点、财务或健康数据等）。若已采取匿名化措施，请描述匿名化流程。 --> 本数据集未包含任何涉密数据，因为其全部由大语言模型（LLM）生成的合成式招聘广告构成，未包含来自个人的真实数据或任何非公开通信内容。此外，生成过程中使用的ESCO分类体系及补充来源数据均已做匿名化处理，不包含任何可识别个人的信息或涉密内容。 本数据集未包含任何可能被视为冒犯、侮辱、威胁或引发焦虑的内容。 由于本数据集全部为由计算方法生成的合成数据，未涉及人类受试者或个人数据，因此无需进行伦理审查。 ## 引用信息（可选） <!-- 若存在介绍该数据集的论文或博客文章，请在此处附上APA与BibTeX格式的引用信息。 --> **BibTeX格式：** [需补充更多信息] **APA格式：** Colombo, S., D'Amico, S., Malandri, L., Mercorio, F., & Seveso, A. (2024). JobSet (1.0.1) [数据集]. Zenodo. https://doi.org/10.5281/zenodo.11484885 ## 数据集卡片联系人 Andrea Seveso - andrea.seveso@unimib.it Antonio Serino - antonio.serino@unimib.it
数据集许可证
- 许可证类型: CC BY 4.0
