clips/beir-nl-quora

BEIR-NL是BEIR基准的荷兰语翻译版本，涵盖了多个领域的信息检索任务，包括事实核查、问答、生物医学信息检索等。数据集包含corpus、queries和qrels文件，分别用于存储文档、查询和查询文档相关性判断。数据集的创建目的是为了促进荷兰语信息检索模型的发展，尽管它是通过自动翻译生成的，可能存在一些局限性，如缺乏原生荷兰语资源、数据污染和基准有效性随时间减弱的问题。

BEIR-NL is a Dutch-translated version of the BEIR benchmark, a diverse and heterogeneous collection of datasets covering various domains from biomedical and financial texts to general web content. BEIR-NL contains tasks such as fact-checking, question-answering, bio-medical IR, argument retrieval, duplicate question retrieval, citation-prediction, and entity retrieval. The dataset structure adheres to the original BEIR benchmark, including a corpus, queries, and relevance judgments file. The dataset was created by automatically translating BEIR datasets into Dutch, with some translations done using GPT-4o-mini and Google Translate. Limitations include reliance on English source data, data contamination issues, and the potential for the benchmarks validity to diminish over time.