基于多特征LSTM-Self-Attention文本情感分类

Text Sentiment Classification Based on Multi-Feature LSTM-Self-Attention

针对自然语言处理情感分析领域中情感分类的问题,提出了一种基于多特征LSTM-Self-Attention的文本情感分类方法。方法以词向量和词性向量为输入,利用LSTM网络模型提取文本的序列特征,并通过在模型中引入自注意力机制(self-attention),从序列特征中提取出句子的语法和语义特征,减少了任务的复杂度。上述方法避免了传统循环神经网络存在的梯度消失和梯度爆炸的问题,极大缩短单词长距离依赖特征之间的距离,提高了分类效果。最后使用中文电影评论数据集进行实验验证,结果表明该方法特征提取能力更强,使得情感分类的准确率提升了1.74%。

Aiming at the problem of sentiment classification in the field of sentiment analysis in natural language processing,a text sentiment classification method based on multi-feature LSTM-Self-Attention was proposed.This method took word vector and part-of-speech vector as input,used the LSTM network model to extract sequence features of text,and introduced a self-attention mechanism into the model to extract grammatical and semantic features of sentences from sequence features,which reduced the complexity of tasks.The method avoided the problem of gradient disappearance and gradient explosion in the traditional Recurrent Neural Network,which greatly shortened the distance between the long-distance dependent features of words and improved the classification effect.Finally,the Chinese film review data set was used for experimental verification.The results show that the feature extraction ability of the method is stronger,and the accuracy of sentiment classification is improved by 1.74%.