摘要
循环神经网络和卷积神经网络能够分别捕捉文本中的长期依赖和局部依赖,但是定长的向量表示限制了循环神经网络的特征表达能力,卷积核的大小也影响了卷积神经网络提取特征的能力。针对这些问题,提出多通道循环卷积神经网络来处理文本分类。采用双向长短期记忆网络对文本进行序列建模;利用标量注意力机制和矢量注意力机制来辅助生成文本的多通道表示;最终由卷积神经网络来完成文本分类。在标准数据集上的实验验证了该框架的分类有效性以及文本多通道表示的语义丰富性。
Recurrent neural network and convolutional neural network can capture long-term and local dependencies in text respectively.However,the fixed-length vector limits the feature expression ability of recurrent neural network,and size of the convolution kernel also affects the feature extraction ability of convolutional neural network.To solve these problems,this paper proposes a multichannel recurrent convolutional neural network for text classification.It used bidirectional long short-term memory network to carry out sequence modeling of text;the scalar attention mechanism and vectorial attention mechanism were used to assist the generation of multichannel representation of text;the final classification task was performed by the convolutional neural network.The experimental results on the benchmark datasets verify the classification effectiveness of the proposed framework and the semantic richness of text multichannel representations.
作者
陆超红
Lu Chaohong(USTC-Birmingham Joint Research Institute in Intelligent Computation and Its Application,School of Computer Science and Technology,University of Science and Technology of China,Hefei 230027,Anhui,China)
出处
《计算机应用与软件》
北大核心
2020年第8期282-288,共7页
Computer Applications and Software
关键词
注意力机制
循环神经网络
卷积神经网络
多通道
文本分类
Attention mechanism
Recurrent neural network
Convolutional neural network
Multichannel
Text classification
