【摘要】 针对文本情感分析中情感极性的问题,提出一种改进型长短期记忆网络模型Senti-LSTM,该模型在现有LSTM网络单元中添加情感门,架构于Senti-LSTM之上的深度神经网络模型充分利用文本上下文信息、文本结构与情感语义等信息,实现更加有效的文本情感表示学习.实验结果表明,Senti-LSTM能够有效提升文本情感分类准确率,同时具有较强的鲁棒性.更多还原

【Abstract】 For sentiment polarity detection in text sequence,an improved long short-term memory unit,called Senti-LSTM,is proposed,which adds a sentiment gate in the classical LSTM.The deep neural network architecture based on Senti-LSTM can more effectively capture sentiment feature of text sequence through full utilization of context,semantic and structure information.Experimental results indicate that,Senti-LSTM can boost performance of sentiment analysis in terms of classification accuracy and robustness.更多还原