FINDSum

FINDSum是首个针对长文本和多表格内容进行摘要的大型数据集，由香港理工大学构建。该数据集基于21,125份来自3,794家公司的年度报告，分为两个子集：FINDSum-ROO关注公司的运营结果，而FINDSum-Liquidity关注公司的流动性。数据集的特点是包含大量的文本和表格数据，以及在目标摘要中包含比以往数据集更多的数值信息。创建过程涉及从公开的EDGAR系统收集HTML格式的10-K表格，并进行文本和表格的预处理。FINDSum的应用领域主要是在金融分析和报告自动化中，旨在提高对复杂金融报告的理解和处理效率。

FINDSum is the first large-scale dataset dedicated to summarizing long texts and multi-table contents, developed by The Hong Kong Polytechnic University. Built upon 21,125 annual reports from 3,794 companies, the dataset is split into two subsets: FINDSum-ROO focuses on corporate operating results, while FINDSum-Liquidity targets corporate liquidity. The dataset features a substantial volume of textual and tabular data, and its target summaries contain more numerical information than those of prior datasets. Its construction involves collecting HTML-formatted 10-K filings from the public EDGAR system, followed by preprocessing of both textual and tabular content. FINDSum is primarily applied in financial analysis and report automation, aiming to improve the efficiency of understanding and processing complex financial reports.

搜集汇总

背景与挑战

背景概述

FINDSum是一个专注于长文本和多表格摘要的大规模数据集，旨在支持相关研究和应用。数据集包含结构化文本和表格数据，通过论文发表于EMNLP 2022和IEEE TKDE 2023，提供Google Drive下载链接并使用ODC-BY许可证。