摘要
新闻和评论文本是进行读者情绪分类的重要资源,但仅仅使用新闻和文本或者把2类文本进行混合作为一组总体特征,不能充分利用不同文本特征间的区别和联系。基于此,提出了一种双通道LSTM(long short-term memory)方法,该方法把2类文本作为2组特征,分别用单通道LSTM神经网络学习这2组特征文本得到文本的LSTM表示,然后通过联合学习的方法学习这2组特征间的关系。实验结果表明,该方法能有效提高读者情绪的分类性能。
The news and comments are important resources to classify the reader emotion. However,previous studies only used news texts or mixed two types of texts as a general feature,which did not make the best use of the differences and connections between different textual features. Based on it,the paper proposed a newapproach named dual-channel LSTM,which treated two types of texts as different features. First,the approach learned a LSTMrepresentation with a LSTMrecurrent neural network. Then,it proposed a joint learning method to learn the relationship between the features. Empirical studies demonstrate the effectiveness of the proposed approach to reader emotion classification.
作者
严倩
王礼敏
李寿山
周国栋
YAN Qian;WANG Li-min;LI Shou-shan;ZHOU Guo-dong(Natural Language Processing Laboratory,Soochow University,Suzhou 215006,Jiangsu,China)
出处
《山东大学学报（理学版）》
CAS
CSCD
北大核心
2018年第9期35-39,48,共6页
Journal of Shandong University(Natural Science)
基金
国家自然科学基金资助项目(61331011
61672366
61375073)
关键词
读者情绪分类
联合学习
双通道LSTM
reader emotion classification
joint learning
dual-channel LSTM
