FIT

FIT数据集是金融领域首个多任务、多模态指令调优数据集，由武汉大学计算机学院等机构创建。该数据集包含136,609条样本，覆盖多种金融任务和数据类型，如情感分析、新闻标题分类、命名实体识别、问答和股票价格预测。数据集的创建过程涉及从公开数据集中收集训练数据，并由领域专家编写特定任务的指令。FIT数据集的应用领域广泛，旨在解决金融文本理解和预测问题，推动金融人工智能的开放源代码发展。The FIT Dataset is the first multi-task and multi-modal instruction-tuning dataset in the financial domain, created by the School of Computer Science of Wuhan University and other institutions. It contains 136,609 samples, covering diverse financial tasks and data types, including sentiment analysis, news headline classification, named entity recognition, question answering, and stock price prediction. The dataset creation process involves collecting training data from public datasets and drafting task-specific instructions by domain experts. The FIT Dataset has a wide range of application scenarios, aiming to solve financial text understanding and prediction problems, and promote the open-source development of financial artificial intelligence.

搜集汇总

背景与挑战

背景概述

FIT is a multi-task and multi-modal financial instruction dataset designed for fine-tuning large language models like FinMA. It includes 136K instruction data samples covering tasks such as sentiment analysis, classification, and stock movement prediction, with both textual and time-series data modalities.