摘要
目前基于卷积神经网络的方法已在情感分类任务中取得了良好的效果。传统的卷积神经网络是将卷积层、池化层及全连接层简单堆积起来的。为了提高卷积神经网络的特征提取能力并加快模型训练速度,对传统的卷积神经网络进行改进,提出分解卷积神经网络模型并将其应用于文本情感分析中。实验结果表明,改进后的卷积神经网络取得了比目前主流的卷积神经网络更好的性能。
Recently,the sentiment classification based on convolutional neural networks have achieved good results.Classic convolutional neural networks is simply stacked by convolutional layers,pooling layers and fully connected layers.For improving the ability of feature extraction and speed up the training of convolutional neural networks,this paper improves the traditional convolutional neural network.This paper proposes the factorize convolutional neural network model and applies it to Chinese sentiment analysis.The experimental results show that the factorize convolution neural networks achieves better performance than basic convolutional neural networks.
作者
孟彩霞
董娅娅
MENG Caixia;DONG Yaya(Xi'an University of Posts and Telecommunications,Xi'an 710000)
出处
《计算机与数字工程》
2019年第8期1970-1973,2101,共5页
Computer & Digital Engineering
基金
陕西省自然科学基金项目（编号：2014JM8303）
陕西省教育厅专项科研计划项目（编号：11JK0988）
西安邮电大学创新基金项目（编号：CXL2015-29）资助
关键词
情感分析
深度学习
特征提取
卷积神经网络
sentiment analysis
deep learning
feature extraction
convolutional neural network
