Event information extraction method and device

According to the event information extraction method and device, event information extraction is carried out through a pre-training model, the training cost can be effectively reduced, good portability is achieved, semantic continuity is guaranteed, and recognition errors are avoided. The method comprises the steps of dividing a target text into a plurality of sub-texts based on continuous semantic information, wherein the sub-texts are used for representing a text containing the continuous semantic information; executing the following process for each sub-text: inputting the sub-text into a trained pre-training model, and outputting entity information related to an event in the sub-text; and determining the event information extracted from the sub-text according to the entity meeting the length requirement in the entity information. 本发明公开了一种事件信息抽取的方法及设备，采用预训练模型进行事件信息抽取，能够有效降低训练成本，具有较好的移植性，还用于保证语义的连续性、避免识别错误。该方法包括：基于连续语义信息将目标文本划分成多个子文本，所述子文本用于表征包含连续语义信息的文本；针对每个子文本执行如下过程：将所述子文本输入到训练好的预训练模型，输出所述子文本中与事件相关的实体信息；根