摘要
针对文本情感分类准确率不高的问题,在卷积神经网络CNN和栈式双向长短时记忆网络B-LSTM的基础上,提出了一种新的情感分析训练模型CNN-BLSTM.该模型利用CNN的卷积操作对词向量进行处理,提取词向量的强度特征,再输入到B-LSTM中进行上层建模,对句子进行处理.结果表明:CNN-B-LSTM模型的情感分类准确率比CNN和B-LSTM模型更高,差错率大约分别降低了4%和1%,具有一定的效果优势.
Aiming at the problem of low accuracy of text sentiment classification, CNN-B-LSTM, a new sentiment analysis training model based on CNN and B-LSTM was presented. The convolution operation processed the word vector to extract the intensity characteristics of the word vector, and then inputed it into the B-LSTM to perform the upper level modeling and used it to process the sentences. The results showed that the proposed CNN-B-LSTM model had higher sentiment classification accuracy, the error rates decreased by 4% and 1% , respectively. It was superior to B-LSTM and CNN models in sentiment classification.
作者
陈欣
于俊洋
赵媛媛
CHEN Xin;YU Junyang;ZHAO Yuanyuan(College of software,He＇nan University,Kaifeng 475000,China;Saier Network Co.,Ltd.,Beijing 100084,China)
出处
《轻工学报》
CAS
2018年第5期103-108,共6页
Journal of Light Industry
基金
河南省科技厅计划发展项目(182102210229)
赛尔网络下一代互联网创新项目(NGII20160204)
