DuEE1.0中文事件抽取数据集

DuEE1.0是百度发布的中文事件抽取数据集，包含65个事件类型的1.7万个具有事件信息的句子（2万个事件）。事件类型根据百度风云榜的热点榜单选取确定，具有较强的代表性。65个事件类型中不仅包含「结婚」、「辞职」、「地震」等传统事件抽取评测中常见的事件类型，还包含了「点赞」等极具时代特征的事件类型。数据集中的句子来自百度信息流资讯文本，相比传统的新闻资讯，文本表达自由度更高，事件抽取的难度也更大。

DuEE 1.0 is a Chinese event extraction dataset released by Baidu. It consists of 17,000 sentences annotated with event information, which collectively represent 20,000 distinct event instances across 65 event categories. These event types were selected based on the trending rankings of Baidu Hot List, thus possessing strong representativeness. The 65 event categories not only cover common event types in traditional event extraction benchmarks, such as 'marriage', 'resignation' and 'earthquake', but also include contemporary event types with distinct era features like 'like'. The sentences in the dataset are sourced from Baidu Feed news texts. Compared with traditional news articles, these texts exhibit higher expressive freedom, which renders event extraction more challenging.