FinQA

该数据集名为FinQA，包含了来自标普500公司财务报告的8,281个财务问题，这些问题涉及在表格和文本数据中生成算术表达式以进行数值推理。在数据分配上，有6,251个问题（占75%）用于训练，883个（占10%）用于开发，以及1,147个（占15%）用于测试。每个问题都附有一个表格，平均包含6.36行数据和24.32个相关句子。该数据集规模为8,281个问题，任务是对表格和文本数据进行的数值推理。

The dataset is named FinQA, which contains 8,281 financial questions sourced from the financial reports of S&P 500 companies. These questions involve generating arithmetic expressions over tabular and textual data to conduct numerical reasoning. For data split, 6,251 questions (accounting for 75% of the total) are used for training, 883 (10%) for development, and 1,147 (15%) for testing. Each question is accompanied by a table, which contains an average of 6.36 rows and 24.32 relevant sentences. The core task of this dataset is numerical reasoning over tabular and textual data.

搜集汇总

背景与挑战

背景概述

FinQA是一个包含8,281个财务问题的数据集，专注于基于表格和文本数据的数值推理任务，问题来源于标普500公司的财务报告，数据集划分为训练、开发和测试三部分，比例为75%、10%和15%。