SALT-NLP/SparkMe-SyntheticUsers

--- license: cc-by-4.0 language: - en tags: - sparkme - synthetic - workforce - interviews - personas task_categories: - text-generation dataset_info: splits: - name: train num_examples: 200 --- # SparkMe-SyntheticUsers Synthetic user profiles for evaluating AI interview systems, released alongside the [SparkMe](https://github.com/SALT-NLP/SparkMe). Each profile represents a simulated workforce participant with demographic metadata, a shuffled list of persona facts, and structured ground-truth interview notes across 10 topics covering the impact of AI in the workplace. ## Dataset Description The 200 profiles were generated from [WorkBank](https://huggingface.co/datasets/SALT-NLP/WORKBank) worker seed data using SparkMe's user agent pipeline. Each user has: - **Demographic metadata** (occupation, gender, education, LLM familiarity, etc.) drawn from the WorkBank seed - **Shuffled facts** (`user_shuffled_facts`) — a flat markdown list of persona facts fed to the LLM user agent during simulated interviews - **Ground-truth notes** (`user_ground_truth_notes`) — structured notes organized by topic and subtopic, used as ground truth for coverage evaluation These profiles are used to evaluate interview system quality via coverage, emergence, and flow metrics. ## Schema | Field | Type | Description | |---|---|---| | `user_id` | string | UUID identifying the user | | `metadata` | dict | Demographic and work background fields | | `user_shuffled_facts` | string | Markdown list of persona facts (shuffled) fed to the user agent | | `user_ground_truth_notes` | list | Structured ground-truth notes across 10 interview topics | ### `user_ground_truth_notes` structure ``` [ { "topic": "Introduction & Background", "subtopics": [ { "subtopic_id": "1.1", "subtopic_description": "...", "notes": ["...", "..."] }, ... ] }, ... ] ``` 10 topics in total (48 subtopics), covering: 1. Introduction & Background 2. Core Responsibilities and Decision-Making 3. Task Proficiency, Challenge, and Engagement 4. Tech Learning Comfort 5. Primary Tools and Technologies Used in Work 6. AI Experience and Tool Adoption 7. AI Interaction Style and Workflow Change 8. Trust and Control Over AI 9. AI Impact on Skills and Job Security 10. AI Attitudes and Future Outlook ## Usage ```python from datasets import load_dataset ds = load_dataset("SALT-NLP/SparkMe-SyntheticUsers") ``` ## Citation If you found our work helpful, please cite our work using the following citation! ```bibtex @article{anugraha2026sparkme, title={SparkMe: Adaptive Semi-Structured Interviewing for Qualitative Insight Discovery}, author={Anugraha, David and Padmakumar, Vishakh and Yang, Diyi}, journal={arXiv preprint arXiv:2602.21136}, year={2026} } ```--- 许可证：CC-BY-4.0 语言：英语 标签：sparkme、合成数据、劳动力、面试、人物画像（personas） 任务类别：文本生成 数据集信息： 划分： - 名称：训练集（train），样本数量：200 --- # SparkMe合成用户数据集（SparkMe-SyntheticUsers） 用于评估AI面试系统的合成用户档案，随[SparkMe](https://github.com/SALT-NLP/SparkMe)一同发布。每份档案均代表一名模拟劳动力参与者，包含人口统计元数据、经打乱排序的人物画像事实列表，以及覆盖职场人工智能影响的10个主题的结构化真实面试笔记。 ## 数据集说明 本数据集的200份档案均基于[WorkBank](https://huggingface.co/datasets/SALT-NLP/WORKBank)的劳动力种子数据，通过SparkMe的用户智能体（user agent）流水线生成。每位用户包含以下信息： - **人口统计元数据**：包含职业、性别、教育程度、大语言模型（LLM）熟悉程度等，均取自WorkBank的种子数据 - **打乱的事实数据**（`user_shuffled_facts`）：经扁平化Markdown列表格式整理的人物画像事实，将在模拟面试过程中输入至大语言模型用户智能体 - **真实标注笔记**（`user_ground_truth_notes`）：按主题与子主题组织的结构化笔记，用作覆盖度评估的真实基准 这些档案可用于通过覆盖度、涌现性与流畅度指标评估面试系统的性能。 ## 数据结构 | 字段 | 类型 | 说明 | |---|---|---| | `user_id` | 字符串 | 用于标识用户的通用唯一识别码（UUID） | | `metadata` | 字典 | 包含人口统计与工作背景的字段 | | `user_shuffled_facts` | 字符串 | 经打乱排序的人物画像事实的Markdown列表，将输入至用户智能体 | | `user_ground_truth_notes` | 列表 | 覆盖10个面试主题的结构化真实标注笔记 | ### `user_ground_truth_notes` 结构 json [ { "topic": "简介与背景", "subtopics": [ { "subtopic_id": "1.1", "subtopic_description": "...", "notes": ["...", "..."] }, ... ] }, ... ] 总计包含10个主题（共48个子主题），覆盖以下内容： 1. 简介与背景 2. 核心职责与决策制定 3. 任务熟练度、挑战与投入度 4. 技术学习接受度 5. 工作中使用的主要工具与技术 6. 人工智能（AI）使用经历与工具采纳情况 7. 人工智能交互风格与工作流程变更 8. 对人工智能的信任与掌控感 9. 人工智能对技能与工作保障的影响 10. 人工智能态度与未来展望 ## 使用方法 python from datasets import load_dataset ds = load_dataset("SALT-NLP/SparkMe-SyntheticUsers") ## 引用方式 若您的工作用到了本数据集，请使用以下引用格式： bibtex @article{anugraha2026sparkme, title={SparkMe: Adaptive Semi-Structured Interviewing for Qualitative Insight Discovery}, author={Anugraha, David and Padmakumar, Vishakh and Yang, Diyi}, journal={arXiv preprint arXiv:2602.21136}, year={2026} }