基于深度双向长短时记忆网络的文本情感分类

Text sentiment classification based on deep bidirectional long-short-term memory network

针对文本情感分类中浅层统计特征忽略了文本内容的序列顺序的问题,提出了一个基于深度双向长短时记忆循环神经网络(DB-LSTM-RNN)的情感分析预测模型。用词嵌入的方法学习文本的分布式表示,并将这些表示作为预训练的向量,用深度双向长短时记忆网络模型进行序列学习,将该结构中学习到的深层表示输入到机器学习分类器中进行情感分类。实验结果表明,该模型比基于浅层统计特征的方法提高了7.6%的准确率。

The shallow statistical feature of text sentiment classification ignores the sequence of text content,so a model for sentiment analysis based on deep bidirectional long-short-term memory recurrent neural network（DB-LSTM-RNN）is proposed.Firstly,a distributed representation of text is learned by word embedding.Then these representations are taken as pre-trained vectors,a deep bidirectional recurrent network with LSTM is used for sequence learning.Finally the deep representations learned from our architecture are fed into a classifier for sentiment classification.Experimental results show that the proposed method can improve the accuracy of 7.6%than the methods based on the statistical characteristics of the shallow layer.