Attention to women's rights in NGO press releases, 1996–2018: A curated, coded dataset of organizational attention to women and violence against women

This dataset contains 31,937 press releases from four leading human rights organizations (Human Rights Watch, Amnesty International, Freedom House, ACLU) spanning 1996-2018, with document-level indicators of attention to women's rights and violence against women.Each document is classified with:• Two binary labels: (1) women/girls depicted as victims; (2) women's/gender rights as main topic• Single best-fitting UDHR category (18 categories)• Country ISO3 code for geographic identificationHuman validation with 200 documents demonstrates strong intercoder reliability (Cohen's κ = 0.848 for women-as-victims, κ = 0.826 for women's-rights-as-topic, κ = 0.783 for UDHR categories, κ = 1.000 for country identification).The dataset enables temporal and cross-national analysis of organizational attention patterns, with straightforward merges to standard country-level datasets (CIRI, PTS, V-Dem) via ISO3 codes.Package Contents:• Core datasets (classified_docs.csv: 31,937 documents; classified_locs.csv: geographic mapping)• Validation materials (intercoder reliability metrics, validation sample, UDHR codebook)• Diagnostic files (unmatched country assignments)• Visualizations of data• Comprehensive README with usage examples in R and PythonAll data are derived from publicly available organizational press releases. The classification pipeline combines LLM-based coding with deterministic fallbacks and rigorous human validation, ensuring valid measurements suitable for cross-national comparative research and temporal analysis of advocacy priorities.

本数据集收录了来自四大顶尖人权组织——人权观察（Human Rights Watch）、大赦国际（Amnesty International）、自由之家（Freedom House）、美国公民自由联盟（ACLU）——1996年至2018年间发布的31937篇新闻稿，附带文档级的女性权利关注度与针对女性暴力的标注指标。
每篇文档均标注以下内容：
• 两项二元标签：(1) 将女性/女孩刻画为受害者；(2) 以女性权利/性别权利为核心主题
• 唯一最贴合的《世界人权宣言》（UDHR）类别（共18类）
• 用于地理标识的国家ISO3代码

针对200篇文档的人工标注验证显示，该数据集具备优异的编码者间信度：女性/女孩作为受害者标签的科恩κ值为0.848，以女性权利为主题标签的κ值为0.826，《世界人权宣言》类别的κ值为0.783，国家标识的κ值为1.000。

本数据集支持针对人权组织关注度模式的时序与跨国分析，可通过ISO3代码与标准国家级数据集（CIRI、PTS、V-Dem）直接关联合并。

数据集包包含以下内容：
• 核心数据集（classified_docs.csv：共31937篇文档；classified_locs.csv：地理映射表）
• 验证材料（编码者间信度指标、验证样本、《世界人权宣言》编码手册）
• 诊断文件（未匹配的国家分配记录）
• 数据可视化内容
• 包含R与Python使用示例的完整README文档

所有数据均源自各组织公开的新闻稿。本数据集的分类流程融合了基于大语言模型（LLM）的编码、确定性回退机制与严格的人工验证，确保所生成的测量结果可适用于跨国比较研究与人权倡导优先级的时序分析。