ibm-research/REAL-MM-RAG_FinSlides_BEIR

这个数据集是BEIR兼容版本的REAL-MM-RAG_FinSlides，包含了2008年至2024年的65个季度财务报告，总共有2280页，主要是表格形式，包含了关键的财务见解。这个数据集的目的是为了评估在视觉结构化财务演示中的检索效果。数据集被分为queries、qrels、docs、corpus和default（queries的别名）五个子集，分别包含了查询信息、查询与文档的相关性、文档信息、图像信息和文档ID。

This dataset is the BEIR-compatible version of REAL-MM-RAG_FinSlides, containing 65 quarterly financial presentations from 2008 to 2024, totaling 2,280 pages, primarily in table format with key financial insights. The purpose of this dataset is to evaluate retrieval in visually structured financial presentations. The dataset is divided into five subsets: queries, qrels, docs, corpus, and default (an alias for queries), which include query information, query-document relevance, document information, image information, and document ID respectively.