摘要
环境和动作是事件的重要特征元素,是抽取特定事件的重要依据。为实现事件抽取,论文提出一种基于BERT、双向门循环控制单元(Bidirectional Gated Recurrent Unit, BiGRU)、条件随机场(CRF)并添加注意力机制的命名实体识别模型用以精确识别文本中的环境和动作要素,经验证,模型对时间要素、地点要素及事件触发词的识别F1值分别达到了0.852 7、0.886 2和0.820 1,具有较好的识别效果。创新性提出事件环境要素匹配算法,用以完成事件-时间-地点的对应划分,从而实现基于特定环境的单一事件抽取。此研究方法可为事件类语料库的构建及应用提供技术参考。
Specific events can be clearly distinguished based on environment and action elements.To achieve event extraction,the named entity recognition model is proposed.The named entity recognition model with BERT,BiGRU,CRF and added attention mechanism can accurately identify the environment and action elements in the text.It is verified that the F1 values of the model for time elements,place elements and event trigger words reach 0.8527,0.8862 and 0.8201,respectively,which has the good recognition effect.An innovative event environment element matching algorithm is proposed to complete the corresponding division of event-time-place,so as to realize single event extraction based on specific environment.This research method can provide technical reference for the construction and application of event corpus.
作者
陈昌佳
代丽
金致楷
王欣悦
CHEN Changjia;DAI Li;JIN Zhikai;WANG Xinyue(School of Economics and Management,Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《智能计算机与应用》
2025年第9期90-95,共6页
Intelligent Computer and Applications
关键词
事件抽取
要素匹配
注意力机制
命名实体识别
event extraction
element matching
attention mechanism
named entity recognition
