摘要
提出一种多粒度门控卷积神经网络(multiple grains-gated convolutional neural networks,MG-GCNN)模型。该模型通过结合词语和单字层面的上下文信息作为网络的输入信息,使网络模型可以充分利用上下文中不同粒度的文本特征信息,并且通过门控操作有效控制不同粒度信息的更新和传递。在不同领域微博文本数据集上的实验结果表明,所提出的MG-GCNN模型取得了比传统分类模型和深度网络模型更好的情感分类效果。
A multiple grains-gated convolutional neural networks(MG-GCNN)model was proposed.By combining the contextual information of words and chars as the input of the network,the proposed model could make full use of the text feature information of different granularities in the context,and effectively control the update and transmission of different granularities information by exploiting the gating operation.The experimental results in different domains of the Chinese microblog text dataset showed that the proposed MG-GCNN model achieved better performance in comparison with the traditional classification models and deep neural network models.
作者
陈珂
梁斌
左敬龙
朱兴统
CHEN Ke;LIANG Bin;ZUO Jinglong;ZHU Xingtong(College of Computer,Guangdong University of Petrochemical Technology,Maoming 525000,China;School of Computer Science and Technology,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China)
出处
《郑州大学学报(理学版)》
CAS
北大核心
2020年第3期21-26,33,共7页
Journal of Zhengzhou University:Natural Science Edition
基金
国家自然科学基金项目(61172145)
广东省自然科学基金项目(2016A030307049,2018A030307032)
广东省高等院校学科与专业建设专项资金项目(2016KTSCX090)。
关键词
中文微博情感分析
门控网络
深度学习
卷积神经网络
自然语言处理
Chinese microblog sentiment analysis
gated network
deep learning
convolutional neural network
natural language processing