<b>DocFEE: A Document-Level Chinese Financial Event Extraction Dataset</b>

<b>DocFEE: A Document-Level Chinese Financial Event Extraction Dataset</b><br><br>DocFEE is a large-scale, document-level dataset designed for financial event extraction in Chinese. It contains ​<b>19,044 annotated documents</b> spanning nine critical financial event types, including <i>Bankruptcy Liquidation</i>, <i>Major Safety Incident</i>, <i>Equity Pledge</i>, and <i>Senior Executive Death</i>, among others. Each document is annotated with ​<b>38 distinct argument types</b> to capture intricate event details, such as dates, amounts, stakeholders, and contextual impacts.The dataset addresses the challenges of document-level event extraction, where events and their arguments may span multiple sentences or exhibit cross-sentence dependencies. On average, documents contain ​<b>1.86 events</b> with a broad event range (<b>960.06</b> <b>characters</b>) and a median document length of<b> 2,277.25</b> <b>characters</b>, reflecting real-world complexity. Key features include fine-grained annotations for diverse financial scenarios, such as tracking shareholder reductions (<i>Reduction Start Date</i>, <i>Shareholder</i>, <i>Reduction Amount</i>) or quantifying losses in <i>Major Asset Loss</i> events (<i>Loss Amount</i>, <i>Other Losses</i>).DocFEE supports research in financial NLP, regulatory compliance, and risk analysis by providing robust, structured data for modeling cross-sentence event relations and argument extraction in long-form texts. Its comprehensive annotations and domain specificity make it a valuable resource for advancing document-level understanding in Chinese financial contexts.<br><br>Please refer to README.pdf for detailed description.

<b>DocFEE：面向中文的文档级金融事件抽取数据集</b><br><br>DocFEE是一款面向中文金融事件抽取任务的大规模文档级数据集。该数据集包含<b>19044份标注文档</b>，涵盖9类核心金融事件类型，包括<i>破产清算（Bankruptcy Liquidation）</i>、<i>重大安全事故（Major Safety Incident）</i>、<i>股权质押（Equity Pledge）</i>以及<i>高管死亡（Senior Executive Death）</i>等。每份文档均标注了<b>38种不同的论元类型</b>，以捕捉事件的细微细节，例如日期、金额、利益相关方以及上下文影响等。该数据集针对文档级事件抽取的挑战进行了优化——在此类任务中，事件及其论元可能跨越多个句子，或存在跨句依赖关系。平均每份文档包含<b>1.86个事件</b>，事件覆盖跨度达<b>960.06字符</b>，文档长度中位数为<b>2277.25字符</b>，充分反映了真实场景的复杂性。其核心特性包括针对多样化金融场景的细粒度标注，例如针对股东减持事件标注<i>减持起始日期（Reduction Start Date）</i>、<i>股东（Shareholder）</i>、<i>减持金额（Reduction Amount）</i>等信息，或针对<i>重大资产损失（Major Asset Loss）</i>事件量化<i>损失金额（Loss Amount）</i>与<i>其他损失（Other Losses）</i>情况。DocFEE通过提供可靠的结构化数据，支持长文本中的跨句事件关系建模与论元抽取任务，可服务于金融自然语言处理（Financial NLP）、监管合规以及风险分析等领域的研究工作。该数据集拥有全面的标注体系与鲜明的领域针对性，是推动中文金融场景下文档级语义理解研究的宝贵资源。<br><br>详细说明请参阅README.pdf文件。