Abhinaya-108/financial-reports-sec

该数据集包含了1993年至2020年间美国上市公司向美国证券交易委员会（SEC）EDGAR系统提交的年度报告（10-K文件）。每个年度报告被划分为20个部分，每个部分进一步拆分为单独的句子。数据集提供了基于申报日期前后市场反应的情感标签，涵盖三个不同时间窗口：[t-1, t+1]、[t-1, t+5]和[t-1, t+30]，标签为积极或消极。此外，数据集还包含了每个申报的元数据，如公司标识（CIK）、句子内容、所属部分、申报日期、文档ID、句子ID等。数据集提供四种配置：large_lite（仅包含基本特征）、large_full（包含所有特征）、small_lite和small_full（分别为精简版和完整版的小规模样本）。该数据集适用于掩码语言建模、情感分析和下一句预测等自然语言处理任务，所有文本均为英文。

The dataset contains the annual report of US public firms filing with the SEC EDGAR system from 1993-2020. Each annual report (10-K filing) is broken into 20 sections. Each section is split into individual sentences. Sentiment labels are provided on a per filing basis from the market reaction around the filing date for 3 different time windows [t-1, t+1], [t-1, t+5] and [t-1, t+30]. Additional metadata for each filing is included in the dataset. Four configurations are available: large_lite (contains only basic features), large_full (includes all features), small_lite and small_full (smaller versions with limited samples). The dataset supports tasks such as masked language modeling, sentiment analysis, and next sentence prediction. All sentences are in English.