摘要
随着知识图谱相关技术的不断发展,垂直领域的事件要素抽取作为知识图谱构建中的重要组成部分受到了学者们的广泛关注。由于垂直领域存在的标注数据极少,采用端到端学习的方法难以取得效果,因此提出了基于卷积神经网络的电力领域事件要素提取方法,该方法采用了电力领域命名的实体识别结果及位置信息的卷积神经网络进行特征提取,并训练多个事件要素判别器完成事件的抽取任务。电力领域数据的结果分析表明,文中所提出的事件要素抽取方法与传统基于依存句法分析的模式匹配方法相比,效果显著。
With the development Knowledge Graph(KG),the event extraction in vertical domain,as an important part of the construction of the KG,has attracted widespread attention by scholars.Since the labeled data in the vertical field is very rare,it is difficult to achieve the significant effect by using the end-to-end learning method.Therefore,in this paper we propose a method to convert the original complex event arguments extraction task into the discrimination of the entity in text.We use Convolutional Neural Networks(CNN)for feature extraction,and we integrate the Named Entity Recognition(NER)result,meanwhile,the position information.Multiple event arguments discriminators are trained for the extraction task.Taking the electricity domain data as an example,the experimental analysis shows that compared with the traditional pattern matching method based on dependency syntax analysis,our methods achieve significant effect.
作者
邓君华
邹云峰
沈盛宇
季梦黎
DENG Junhua;ZOU Yunfeng;SHEN Shengyu;JI Mengli(Marketing Service Center,State Grid Jiangsu Electric Power Co.,Ltd.,Nanjing 210019,China;Nanjing Yunwen Network Technology Co.,Ltd.,Nanjing 211000,China)
出处
《电子设计工程》
2021年第3期132-135,140,共5页
Electronic Design Engineering
基金
国网江苏省电力有限公司科技项目(J2018020)。
关键词
事件要素抽取
知识图谱
电力领域
卷积神经网络
特征提取
event factor extraction
Knowledge Graph
power field
Convolutional Neural Network
feature extraction
