oliverkinch/tidsskrift-dk-bt

tidsskrift.dk回译数据集是一个基于丹麦学术期刊文章的指令回译数据集。该数据集通过从人文、社会科学、自然科学和专业领域的丹麦期刊中采样文章，并使用指令生成模型为每篇文章生成合成用户提示，从而构建了用户提示与期刊原文配对的训练数据。数据集包含多个字段，如唯一标识符(id)、合成用户提示(prompt)、原始期刊文本(target)以及来源元数据(sources)等。此外，数据集还提供了详细的期刊条目数量统计，涵盖了多个不同领域的期刊。

The tidsskrift.dk Backtranslation dataset is an instruction backtranslation dataset derived from Danish academic journal articles. It pairs synthetic user prompts (generated by an LLM) with the original journal article texts as target responses. The dataset was constructed by sampling articles from Danish journals spanning humanities, social sciences, natural sciences, and professional fields, and generating diverse prompts of varying lengths. It includes fields such as unique row identifier (id), synthetic user prompt (prompt), original journal text (target), and source metadata (sources). The dataset also provides detailed journal counts across different fields.