sw4tanonymous/FinAR-Bench

FinAR-Bench数据集是一个用于评估大型语言模型在执行金融基本面分析方面的能力的数据集。它包含三个关键任务：信息提取、指标计算和逻辑推理。数据集由以下部分组成：1. 公司的年度报告的财务报表部分（PDF格式），这些报告来自2023年上海证券交易所上市的100家公司；2. 使用六种不同的PDF提取工具从PDF文档中提取的文本；3. 开发集和测试集，每个集合包含10家和90家公司的评估任务，分别包含事实提取任务、财务指标计算任务和逻辑推理任务。数据集的结构包括JSON格式的评估数据、原始的PDF文件和从PDF提取的文本文件。

The FinAR-Bench dataset is designed to assess the capabilities of Large Language Models (LLMs) in performing financial fundamental analysis. It focuses on three key tasks: Information Extraction, Indicator Computation, and Logical Reasoning. The dataset consists of the following components: 1. Financial statements in PDF format from the 2023 annual reports of 100 companies listed on the Shanghai Stock Exchange (SSE); 2. Text extracted from these PDF documents using six different PDF extraction tools; 3. Development and test sets with evaluation tasks for 10 and 90 companies respectively, including fact extraction tasks, financial indicator computation tasks, and logical reasoning tasks. The dataset structure includes JSON-formatted evaluation data, original PDF files, and text files extracted from the PDFs.