摘要
CNN在处理短文本情感分类时,使用卷积层抽取局部特征,用最大池化层选取局部特征最大值,易忽略其长期序列特性。该文使用一种新的深度学习模型ConvLSTM,利用长LSTM替代CNN中的最大池化层,以减少局部信息的丢失并捕获句子序列中的长期依赖关系。在IMDB影评数据集和Amazon评论数据集上的实验表明,该模型较CNN和单纯的LSTM在准确率、召回率和F值等方面均有较明显的提高。
When CNN processes emotion classification of short texts,it extracts local features by using convolutional layer and selects the maximum value of local features by using maximum pooling layer,which is easy to ignore its long term sequence characteristics.A new deep learning model ConvLSTM is used to replace the maximum pooling layer in CNN with long LSTM,so as to reduce the loss of local information and capture the long term dependence in sentence sequence.Experiments on IMDB film review data set and Amazon review data set show that the accuracy,recall rate and F value of the model are significantly higher than those of CNN and LSTM.
作者
张志远
万双双
ZHANG Zhiyuan;WAN Shuangshuang(College of Computer Science&Technology,Civil Aviation University of China,Tianjin 300300,China)
出处
《现代电子技术》
北大核心
2019年第22期159-163,共5页
Modern Electronics Technique
基金
国家自然科学基金民航联合基金项目(U1633110)
中央高校基本科研业务费专项基金资助项目(3122016D021)~~
