摘要
关系抽取是信息抽取中的一个重要子任务,很多关系抽取任务利用现有的词法分析和句法分析等基本的NLP处理工具来生成特征,但是该特征提取方法完全利用之前的经验进行,特征的质量过度依赖于已有的NLP工具的准确率,存在误差传播的问题。近年来随着深度学习的发展,卷积神经网络学习算法在很多自然语言处理任务中取得了较好的效果。文中应用计算机领域的手工标注数据集,采用卷积神经网络的实体关系抽取方法。实验表明,卷积神经网络方法能够有效抽取实体之间的关系,其准确率和F1值较BiLSTM方法有所提高。
Relationship extraction is an important sub-task in information extraction.In many relational extraction tasks,the existing lexical analysis,syntax analysis and other basic NLP processing tools are used to generate features.However,the feature extraction method is carried out completely based on previous experience.The quality of features is much too dependent on the accuracy of existing NLP tools,and there is a problem of error propagation.With the development of deep learning in recent years,convolutional neural network learning algorithm has achieved good results in many natural language processing tasks.In this paper,the method of manually annotated data set in computing field is adopted,and the entity relation extraction method of the convolutional neural network is used.The experiment results show that convolutional neural network can extract the relationship between entities effectively,and its accuracy and F1 value are improved compared with the BiLSTM method.
作者
排日旦·阿布都热依木
吐尔地·托合提
艾斯卡尔·艾木都拉
Peride ABDUREHIM;Turdy TOHTI;Askar HAMDULLA(College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China)
出处
《信息技术》
2022年第3期1-5,11,共6页
Information Technology
基金
国家重点研发计划项目(2017YFC0820603)。
关键词
关系抽取
自然语言处理
特征提取
深度学习
卷积神经网络
relation extraction
natural language processing
feature extraction
deep learning
convolutional neural network
