sionic-ai/NanoBEIR-ja

NanoBEIR-ja是一个日语信息检索评估基准数据集，包含预处理过的查询，属于NanoBEIR基准的一部分。数据集分为三个配置：corpus、qrels和queries，每个配置包含多个子集，如NanoClimateFEVER、NanoDBPedia等。该数据集专为文本检索任务设计，标记为sentence-transformers和retrieval。预处理流程包括格式检测、转换以及使用Gemini 2.5 Flash和GPT-4o模型进行质量验证。提供了预处理查询的示例，展示了从原始陈述到问题格式的转换。

NanoBEIR-ja is a Japanese benchmark dataset for information retrieval evaluation with preprocessed queries, part of the NanoBEIR benchmark. The dataset is structured into three configurations: corpus, qrels, and queries, each with multiple splits corresponding to subsets like NanoClimateFEVER, NanoDBPedia, etc. It is designed for text retrieval tasks and tagged with sentence-transformers and retrieval. The preprocessing pipeline involves format detection, conversion, and quality validation using models like Gemini 2.5 Flash and GPT-4o. Examples of preprocessed queries are provided, demonstrating the transformation from original statements to question formats.