mteb/Vidore2EconomicsReportsRetrieval

这是一个多语言数据集，旨在用于视觉文档检索、图像到文本和文本到图像的任务。数据集包括来自源数据集vidore/economics_reports_v2的派生注释，并支持MTEB（大规模文本嵌入基准）基准。该数据集在四种语言中可用：德语（deu）、英语（eng）、法语（fra）和西班牙语（spa）。它遵循MIT许可证，并包含三种主要类型的数据：语料库、查询和qrels（查询相关性分数）。每种类型的数据都分为“测试”集。数据集针对每种语言有不同的配置，每个配置都有自己的特征集，例如“图像”、“文档-id”、“id”、“文本”、“查询-id”、“语料库-id”和“分数”。该数据集旨在用于评估嵌入模型，并是MTEB项目的一部分，该项目专注于跨多种语言和任务对文本嵌入模型进行基准测试。

This is a multilingual dataset designed for visual-document retrieval, image-to-text, and text-to-image tasks. It includes annotations derived from the source dataset vidore/economics_reports_v2 and is compatible with the MTEB (Massive Text Embedding Benchmark) benchmark. The dataset is available in four languages: German (deu), English (eng), French (fra), and Spanish (spa). It is licensed under the MIT license and consists of three main types of data: corpus, queries, and qrels (query relevance scores). Each type of data is split into a test set. The dataset is structured with different configurations for each language, each with its own set of features, such as image, doc-id, id, text, query-id, corpus-id, and score. It is intended for evaluating embedding models and is part of the MTEB project, which focuses on benchmarking text embedding models across multiple languages and tasks.