WikiOmnia

WikiOmnia是由独立研究员和人工智能研究所（AIRI）创建的公开可用QA数据集，包含与俄语维基百科文章摘要部分对应的QA对。该数据集通过完全自动化的生成和过滤流程构建，确保了生成QA对的高质量。WikiOmnia不仅适用于俄语，其自动化流程还测试了在新闻、小说和社会媒体等其他领域的应用。数据集分为两部分：原始数据包含近1600万对QA，而经过严格自动验证的清洗数据则包含超过16万对QA。WikiOmnia旨在通过提供大规模、高质量的QA数据，推动机器阅读理解系统的发展，特别是在信息检索和多语言应用领域。

WikiOmnia is a publicly available QA dataset created by independent researchers and the Artificial Intelligence Research Institute (AIRI). It consists of QA pairs corresponding to the abstract sections of Russian Wikipedia articles. This dataset is constructed through a fully automated generation and filtering pipeline, ensuring the high quality of the generated QA pairs. Beyond supporting Russian language applications, its automated pipeline has also been tested for use in other domains such as news, fiction, and social media. The dataset is divided into two subsets: the raw data contains nearly 16 million QA pairs, while the rigorously automatically validated cleaned subset includes over 160,000 QA pairs. WikiOmnia aims to advance the development of machine reading comprehension systems, particularly in the fields of information retrieval and multilingual applications, by providing large-scale, high-quality QA data.