DocFinQA
收藏资源简介:
DocFinQA是一个专注于金融领域长文档问答任务的数据集,由Kensho Technologies创建。该数据集扩展了现有的FinQA数据集,增加了7,437个问题,并将平均上下文长度从FinQA的不到700字扩展到DocFinQA的123,000字。数据集中的每个问题都与完整的文档上下文相关联,旨在更真实地模拟金融专业人士处理数百页文档时的情况。DocFinQA不仅用于评估模型在处理长文档时的推理能力,还特别关注金融领域的数值推理,为金融分析、基因序列分析和法律文档合同分析等领域的模型提供了挑战和改进的机会。
DocFinQA is a dataset focused on the long-document question answering task within the financial domain, created by Kensho Technologies. This dataset expands on the existing FinQA dataset by adding 7,437 questions, and extends the average context length from under 700 words in FinQA to 123,000 words in DocFinQA. Each question in the dataset is associated with full document context, aiming to more realistically simulate the scenario where financial professionals process hundreds of pages of documents. DocFinQA is not only used to evaluate the reasoning ability of models when processing long documents, but also pays special attention to numerical reasoning in the financial field, providing challenges and improvement opportunities for models in fields such as financial analysis, gene sequence analysis, and legal document contract analysis.

- 1DocFinQA: A Long-Context Financial Reasoning DatasetKensho Technologies · 2024年



