摘要
目前基于深度学习的人物关系抽取方法中大都使用单一的CNN或RNN模型,CNN擅长提取局部的最重要特征,但是不适合处理序列输入,而RNN虽然在任意长度的序列化任务中具有很大的优势,但是对局部的重要特征提取不够充分,针对以上问题,提出一种基于双向GRU和PCNN的人物关系抽取方法。将双向GRU模型与PCNN模型的优点结合起来,同时在模型中加入注意力机制。利用远程监督的方法构建训练语料进行实验验证,结果表明,该方法相较于单一的双向GRU模型、PCNN模型具有更好的效果。
At present,most of the character relationship extraction methods based on deep learning use a single CNN or RNN model.CNN is good at extracting the most important local feature,but it’s not suitable for processing the sequence input,while RNN has great advantages in the serialization task of arbitrary length,but the extraction of the most important local feature is not enough.To solve the above problems,a method of character relationship extraction based on bidirectional GRU and PCNN is proposed.Combine the advantages of Bi⁃GRU and PCNN,add the attention mechanisms into the model.The method of remote supervision is used to construct the training corpus for experimental verification.The results show that this method has better effect than the single bidirectional GRU model and PCNN model.
作者
王明波
王峥
邱秀连
WANG Ming⁃bo;WANG Zheng;QIU Xiu⁃lian(Wuhan Research Institute of Posts and Telecommunications,Wuhan 430000,China;Nanjing FiberHome World Communication Technology Co.Ltd.,Nanjing 210000,China)
出处
《电子设计工程》
2020年第10期160-165,共6页
Electronic Design Engineering
关键词
深度学习
人物关系抽取
双向门循环单元
卷积神经网络
deep learning
character relationship extraction
bidirectional gated recurrent unit
convolu tional neural network
