摘要
提出一种基于关键n-grams和门控循环神经网络的文本分类模型.模型采用更为简单高效的池化层替代传统的卷积层来提取关键的n-grams作为重要语义特征,同时构建双向门控循环单元(gated recurrent unit,GRU)获取输入文本的全局依赖特征,最后将两种特征的融合模型应用于文本分类任务.在多个公开数据集上评估模型的质量,包括情感分类和主题分类.与传统模型的实验对比结果表明:所提出的文本分类模型可有效改进文本分类的性能,在语料库20newsgroup上准确率提高约1.95%,在语料库Rotton Tomatoes上准确率提高约1.55%.
An effective text classification model based on n-grams and a gated recurrent neural network is proposed in this paper.First,we adopt a simpler and more efficient pooling layer to replace the traditional convolutional layer to extract the essential n-grams as important semantic features.Second,a bidirectional gated recurrent unit(GRU)is constructed to obtain the global dependency features of the input text.Finally,we apply the fusion model of the two features to the text classification task.We evaluate the quality of our model on sentiment and topic categorization tasks over multiple public datasets.Experimental results show that the proposed method can improve text classification effectiveness compared with the traditional model.On accuracy,it approaches an improvement of 1.95%on the 20 newsgroup and 1.55%on the Rotten Tomatoes corpus.
作者
赵倩
吴悦
刘宗田
ZHAO Qian;WU Yue;LIU Zongtian(School of Computer Engineering and Science,Shanghai University,Shanghai 200444,China)
出处
《上海大学学报（自然科学版）》
CAS
CSCD
北大核心
2021年第3期544-552,共9页
Journal of Shanghai University:Natural Science Edition
