摘要
对命名实体之间的语义关系抽取进行研究。分别使用Word2vec和GloVe对句子最短依存路径进行词向量表示,作为卷积神经网络和双向长短期记忆网络的输入,通过网络自动学习特征,通过拼接的方式将两种特征融合,通过softmax分类器得出所属关系的类型。采用SemEval-2010 Task 8数据集,实验结果表明,使用多种词向量表示最短依存路径和通过拼接的方式融合卷积神经网络与长短期记忆网络的特征能显著提高关系抽取的效果。
Semantic relation extraction between named entities was researched.Word2vec and GloVe were respectively used to represent sentence’s shortest dependency path,which were taken as the inputs of convolutional neural network and bi-directional long-short term memory network,then features were automatically learned through the network and these two kinds features were concatenated.The type of relation was concluded through softmax classifier.SemEval-2010 Task 8 dataset was used.Experimental results show that the results of relation extraction are significantly improved by using shortest dependency path represented by multiple word embedding and concatenating method of convolutional neural network and long-short term memory network’s feature.
作者
温政
段利国
李爱萍
WEN Zheng;DUAN Li-guo;LI Ai-ping(College of Information and Computer,Taiyuan University of Technology,Taiyuan 030024,China;State Key Laboratory of Software Engineering,Wuhan University,Wuhan 430072,China)
出处
《计算机工程与设计》
北大核心
2019年第9期2672-2676,2696,共6页
Computer Engineering and Design
基金
山西省自然科学基金项目(2013011015-2)
武汉大学软件工程国家重点实验室开放课题基金项目(SKLSE2012-09-30)
关键词
关系抽取
最短依存路径
双通道
卷积神经网络
双向长短期记忆网络
relation extraction
shortest dependency path
bi-channel
convolutional neural network
bi-directional long-short term memory network
