trading-data
收藏资源简介:
DartLab数据集是一个结构化的公司披露数据集合,源自韩国金融监督院的DART(电子披露系统)和美国证券交易委员会的EDGAR(电子数据收集、分析与检索系统)。该数据集作为DartLab Python库的数据层,旨在将公开的公司披露文件转化为统一、结构化的公司信息图谱。数据集包含约2,700家韩国公司和970家美国公司的数据,总计涵盖近3,000家公司。数据以Parquet文件格式组织,每个文件对应一家公司(通过股票代码标识)。数据集分为三个主要子集:1) **docs**:包含年度/季度报告中的全文章节,解析为结构化的文本块(如文本、表格、标题),涵盖约2,547家公司,数据量约8 GB。关键字段包括披露ID、申报日期、股票代码、公司名称、报告类型、章节标题、内容(Markdown格式)、区块类型和年份。2) **finance**:包含基于XBRL的财务报表数据(如资产负债表、损益表、现金流量表),源自DART OpenAPI,涵盖约2,744家公司,数据量约586 MB。关键字段包括业务年份、报告季度代码、财务报表类型(合并或单独)、科目ID、科目名称(韩语)以及当期、前期和更早期间的金额。3) **report**:包含28种DART API类别的结构化披露数据,涉及公司治理、薪酬、持股等方面,涵盖约2,711家公司,数据量约319 MB。关键字段包括API类型(如股息、员工、高管)、年份、季度、股票代码和公司代码,以及各类别特定的列。数据集适用于表格问答、文本分类、财务分析、公司信息披露研究等任务,支持韩语和英语。数据每日通过GitHub Actions自动更新,确保包含最近7天内的新披露文件。所有数据均保持原始申报文件中的数值,未经四舍五入、估算或插值处理。数据集采用Apache 2.0许可证。
The DartLab dataset is a structured collection of corporate disclosure data, sourced from DART (Data Analysis, Retrieval and Transfer System) operated by the Korean Financial Supervisory Service and EDGAR (Electronic Data Gathering, Analysis, and Retrieval) by the U.S. Securities and Exchange Commission. It serves as the data layer for the DartLab Python library, aiming to transform public corporate disclosure documents into a unified, structured corporate information graph. The dataset includes data from approximately 2,700 Korean companies and 970 U.S. companies, totaling nearly 3,000 companies. Data is organized in Parquet file format, with each file corresponding to a company (identified by stock ticker). The dataset is divided into three main subsets: 1) **docs**: Contains full-text sections from annual/quarterly reports, parsed into structured text blocks (e.g., text, tables, headings), covering about 2,547 companies with a data volume of approximately 8 GB. Key fields include disclosure ID, filing date, stock ticker, company name, report type, section title, content (in Markdown format), block type, and year. 2) **finance**: Includes XBRL-based financial statement data (e.g., balance sheets, income statements, cash flow statements), sourced from DART OpenAPI, covering about 2,744 companies with a data volume of approximately 586 MB. Key fields include business year, report quarter code, financial statement type (consolidated or separate), account ID, account name (in Korean), and amounts for the current, previous, and earlier periods. 3) **report**: Contains structured disclosure data from 28 DART API categories, covering areas such as corporate governance, compensation, and shareholding, with about 2,711 companies and a data volume of approximately 319 MB. Key fields include API type (e.g., dividend, employee, executive), year, quarter, stock ticker, company code, and category-specific columns. The dataset is suitable for tasks such as table question answering, text classification, financial analysis, and corporate disclosure research, supporting both Korean and English. Data is automatically updated daily via GitHub Actions to include new disclosure files from the past 7 days. All data retains the original values from the filing documents, without rounding, estimation, or interpolation. The dataset is licensed under Apache 2.0.
数据集概述:DartLab 전자공시 데이터 (DartLab Data)
该数据集由 DartLab 库提供支持,是经过预处理的结构化企业披露数据,涵盖韩国 (DART) 和美国 (EDGAR) 的上市公司信息。数据集以 Apache Parquet 格式存储。
- 任务类型:表格问答、文本分类
- 语言:韩语、英语
- 许可证:Apache-2.0
- 数据规模:1,000 到 10,000 条记录
数据来源
- 韩国 DART:韩国金融监督院运营的电子披露系统 (dart.fss.or.kr)
- 美国 EDGAR:美国证券交易委员会的电子数据收集分析和检索系统 (sec.gov/edgar)
- 所有数据均来自公开的政府披露系统,财务数据保持原始申报文件原样,未做四舍五入、估算或插值。
数据结构
数据集分为三个主要子集,每个文件代表一家公司,文件名为 {股票代码}.parquet。
| 子集名称 | 覆盖公司数 | 数据大小 | 描述 |
|---|---|---|---|
dart/docs/ |
2,547 家 | 约 8 GB | 年度/季度报告全文,解析为结构化区块 (文本、表格、标题) |
dart/finance/ |
2,744 家 | 约 586 MB | 基于 XBRL 的财务报表(资产负债表、损益表、现金流量表等) |
dart/report/ |
2,711 家 | 约 319 MB | 28 种 DART 结构化披露 API 数据(治理、薪酬、股权等) |
docs子集列:rcept_no(申报编号),rcept_date(申报日期),stock_code(股票代码),corp_name(公司名),report_type(报告类型),section_title(章节标题),section_order(章节顺序),content(章节文本),blockType(区块类型),year(年份)finance子集列:bsns_year(业务年度),reprt_code(报告季度代码),stock_code(股票代码),corp_name(公司名),fs_div(财务报告类别:合并/个别),sj_div(报表类型),account_id(XBRL 账户 ID),account_nm(账户名,韩语),thstrm_amount(本期金额),frmtrm_amount(上期金额),bfefrmtrm_amount(前两期金额)report子集包含 28 种 API 类型,包括:dividend(股息),employee(员工),executive(高管),majorHolder(大股东),treasuryStock(库存股),capitalChange(资本变动),auditOpinion(审计意见),stockTotal(股票总数),outsideDirector(外部董事),corporateBond(公司债券) 等
更新计划
通过 GitHub Actions 每日自动检查并增量收集最近 7 天的申报文件。




