FAMMA
收藏资源简介:
FAMMA是一个开源的金融多语言多模态问答基准数据集,由浙江大学和Ant Group共同创建。该数据集包含1758个精心收集的问题-答案对,涵盖了公司金融、资产管理、金融工程等8个主要金融子领域。数据集中的问题以中、英、法三种语言呈现,并结合了文本和异构图像类型,如图表、表格和图示。创建过程中,数据集通过两阶段质量控制确保数据准确性,并根据CFA课程标准进行难度分类。FAMMA旨在评估多模态大语言模型在复杂金融知识问答中的能力,推动金融领域专家系统的研究。
FAMMA is an open-source financial multilingual multimodal question answering benchmark dataset co-created by Zhejiang University and Ant Group. This dataset contains 1758 meticulously collected question-answer pairs, covering 8 major financial subfields such as corporate finance, asset management, financial engineering, and others. The questions in the dataset are presented in three languages: Chinese, English and French, and combine text and heterogeneous image types including charts, tables and diagrams. During the creation process, the dataset adopted a two-stage quality control mechanism to ensure data accuracy, and the difficulty of each sample was classified according to the CFA curriculum standards. FAMMA is designed to evaluate the capabilities of multimodal large language models (LLMs) in complex financial knowledge-based question answering tasks, and advance the research of expert systems in the financial domain.

- 1FAMMA: A Benchmark for Financial Domain Multilingual Multimodal Question Answering浙江大学 · 2024年



