KILT Wikipedia

该数据集是通过将维基百科文章分割成100个单词的片段（对于非空白分隔的语言，则是100个Unicode字符），并在每个片段前加上文章标题来构建的。这种方法特别适用于多语言环境下的检索增强生成任务，该数据集已被用于多项语言的检索实验研究中。This dataset is constructed by splitting Wikipedia articles into 100-word segments (or 100 Unicode characters for non-whitespace-separated languages), with each segment prepended with the article's title. This approach is particularly suitable for retrieval-augmented generation (RAG) tasks in multilingual environments. This dataset has been used in retrieval experimental studies across multiple languages.