FinQA, DM-Simplong, XBRL-Math
收藏资源简介:
本文所使用的数据集包括FinQA、DM-Simplong和XBRL-Math三个数据集,分别用于评估模型在金融文本、表格数据和公式方面的数值推理、表格解释、金融术语理解、长文本处理和基于公式的解决问题能力。FinQA是一个大规模的金融推理数据集,专注于复杂的数值推理;DM-Simplong专为评估长文本中的数值推理设计;XBRL-Math则用于评估模型在XBRL财务报告中的数值推理能力。这些数据集涵盖了金融领域特有的挑战,如理解金融术语、从不同来源的财务报告中提取相关数字和实体,以及处理长文本和多表格。
The datasets utilized in this paper consist of three benchmark datasets: FinQA, DM-Simplong, and XBRL-Math. These datasets are designed to evaluate models' core capabilities including numerical reasoning, table interpretation, financial terminology comprehension, long-text processing, and formula-driven problem-solving across financial texts, tabular data, and mathematical scenarios, with each dataset targeting a specific evaluation focus. Specifically, FinQA is a large-scale financial reasoning dataset dedicated to complex numerical reasoning; DM-Simplong is specially developed to assess numerical reasoning in long-form texts; and XBRL-Math is employed to evaluate models' numerical reasoning abilities within XBRL financial reports. These datasets encompass the unique challenges inherent in the financial domain, such as understanding financial terminology, extracting relevant numbers and entities from financial reports of diverse sources, and processing long texts and multiple tables.

- 1Fino1: On the Transferability of Reasoning Enhanced LLMs to FinanceTheFinAI · 2025年



