MT-RAIG BENCH
收藏资源简介:
MT-RAIG BENCH是由延世大学DLI实验室创建的大型基准数据集,旨在评估系统在多个表格上的检索增强型洞见生成能力。该数据集包含18532个测试示例,涉及19563个独特的表格,覆盖了5418个独特的表格集。每个示例平均涉及2.88个金标表格。数据集通过自动化和人工审核相结合的方式构建,包含分析总结、比较关系、性能结果和趋势模式等多种类型的问题。
MT-RAIG BENCH is a large-scale benchmark dataset created by the DLI Lab at Yonsei University, aiming to evaluate the retrieval-augmented insight generation capabilities of systems across multiple tables. This dataset contains 18,532 test examples, involving 19,563 unique tables and covering 5,418 unique table collections. On average, each example refers to 2.88 gold-standard tables. The dataset is constructed through a hybrid approach combining automation and manual review, and includes multiple types of questions such as analytical summaries, comparative relationships, performance results and trend patterns.

- 1MT-RAIG: Novel Benchmark and Evaluation Framework for Retrieval-Augmented Insight Generation over Multiple Tables延世大学 · 2025年



