<b>Dataset for "Generative AI and the Boundaries of Human Judgment" (Chinese HSS Reviewers）</b>

This dataset supports the manuscript <b>“Generative AI and the Boundaries of Human Judgment: A Study of Chinese HSS Reviewers in International Open‑Access Journals”</b>. It contains richly annotated qualitative data illustrating how veteran Chinese humanities and social sciences (HSS) reviewers use generative AI tools during peer review.📄 Data Overview<b>10 semi‑structured interviews</b> (R1–R10), conducted in Mandarin between June 1–30, 2025<b>Verbatim transcripts</b>, back-translated into English for cross-cultural analysis<b>Codebook (Appendix A)</b>: 202 initial codes linked to 4 core themes—Patterns of AI Use, Human Oversight &amp; Quality Control, Disclosure &amp; Governance, Ethical &amp; Privacy Concerns<b>Coding-trail memo (Appendix B)</b>: shows how codes were clustered into subthemes and then into final themes using Braun &amp; Clarke’s six-phase framework🛠️ Methods &amp; EthicsParticipants: Experienced Chinese HSS peer reviewers with ≥10 international open-access reviewsData collection: Audio-recorded interviews, anonymized, IRB-approved, encrypted local storageAnalysis: Dual independent coding using NVivo 12 (Cohen’s κ = 0.82); thematic saturation reached by interview 10Verification: Triangulation across career stage and discipline to improve internal validity🔍 Research AimsInvestigate which review tasks prompt AI use (grammar, citations, structure, visualization)Understand layered verification habits—cross-platform checks, source tracing, expert consultationAnalyze how reviewers navigate cultural nuance, confidentiality, and disclosure given vague journal policies🌱 Reuse PotentialSupports replication in other cultural or disciplinary contextsEnables comparative studies of AI-assisted peer review practicesProvides a rich codebook for secondary thematic or discourse analysis📁 File Structure<pre>bash复制编辑<pre>/transcripts/<br> R1_Mandarin.txt <br> R1_English.txt <br> … (R1–R10) <br>/codebook.xlsx <br>/memo_coding_trail.pdf<br></pre></pre>Transcripts labeled R1–R10, each with Mandarin and English versions<code>codebook.xlsx</code>: lists 202 codes, subthemes, corresponding interview IDs<code>memo_coding_trail.pdf</code>: documents thematic mapping process (Appendix B)🔐 Ethics &amp; LicensingFully anonymized: no personal identifiers includedData collection followed IRB guidelines under [University Name]Licensed under <b>CC BY 4.0</b>Complies with Figshare’s human-subject data policies🔗 Related Resources<b>Article DOI:</b> [to be populated upon acceptance]<b>Supplementary materials</b> (Appendix A &amp; B) available via journal repository

本数据集支持论文《生成式AI (Generative AI) 与人类判断的边界：国际开放获取期刊中的中国人文社科（Humanities and Social Sciences, HSS）审稿人研究》。本数据集包含标注详尽的质性数据，用以展现中国资深人文社科审稿人在同行评审过程中如何使用生成式AI工具。

📄 数据概览
1. 10次半结构化访谈（编号R1–R10），于2025年6月1日至30日以普通话开展；
2. 逐字访谈转录稿，经反向翻译为英文以支撑跨文化分析；
3. 编码手册（附录A）：包含202个初始编码，归属于4个核心主题——AI使用模式、人类监督与质量控制、披露与治理、伦理与隐私关切；
4. 编码追踪备忘录（附录B）：展示了如何借助Braun & Clarke六阶段框架，将编码聚类为子主题，最终整合为核心主题。

🛠️ 方法与伦理
研究对象：拥有≥10次国际开放获取期刊审稿经验的中国资深人文社科同行评审专家。
数据采集：采用录音访谈，已完成匿名化处理，通过伦理审查委员会（IRB）批准，存储于加密本地存储介质。
数据分析：采用NVivo 12软件开展双独立编码，科恩κ系数（Cohen’s κ）为0.82；至第10次访谈时达成主题饱和。
验证方法：通过审稿人职业阶段与学科背景的三角验证法提升内部效度。

🔍 研究目标
1. 探究哪些审稿任务会触发AI工具的使用（如语法校对、参考文献整理、结构优化、可视化制作）；
2. 了解审稿人的分层验证习惯——跨平台核查、来源溯源、专家咨询；
3. 分析在期刊政策模糊的情况下，审稿人如何应对文化差异、保密要求与披露规范。

🌱 复用潜力
本数据集可支持其他文化或学科语境下的研究复现，可用于开展AI辅助同行评审实践的比较研究，同时提供了丰富的编码手册以供二次主题分析或话语分析使用。

📁 文件结构
bash
/transcripts/
R1_Mandarin.txt
R1_English.txt
…（覆盖R1–R10全部样本）
/codebook.xlsx
/memo_coding_trail.pdf

其中，转录稿以R1–R10编号，每份均包含普通话原文与英文译本；`codebook.xlsx` 列明202个编码、子主题及对应访谈编号；`memo_coding_trail.pdf` 记录了主题映射流程（即附录B内容）。

🔐 伦理与许可
本数据集已完成完全匿名化处理，未包含任何个人身份识别信息；数据采集严格遵循[机构名称]下属伦理审查委员会的指南；采用CC BY 4.0许可协议发布；符合Figshare平台的人类受试者数据政策要求。

🔗 相关资源
- 论文DOI：[待录用后补充]
- 附录A与附录B等补充材料可通过期刊知识库获取。