基于深度学习的文本情感分析

Text sentiment analysis based on deep learning

传统的文本情感分析主要基于情感词典、机器学习以及传统的神经网络模型等实现特征的提取及情感的分类,但由于语料简短及特征稀疏,使得这类情感分析方法取得的效果不理想.因此,提出采用基于Self-Attention机制的卷积神经网络(CNN)和双向长短时记忆网络(BI-LSTM)相结合的模型结构(SCBILSTM模型)对微博文本进行情感分析,SCBILSTM利用双向循环神经网络对文本上下文进行特征提取,并利用CNN进行局部特征提取,在此基础上添加自注意力机制,在通过网络爬虫抓取的微博数据集上和其他模型进行对比实验,验证本文中所提出的模型有效提升了文本分类的准确率.

The traditional text sentiment analysis is mainly based on emotion dictionary,machine learning and traditional neural network model to achieve feature extraction and emotion classification.However,due to the short corpus and sparse features,this kind of sentiment analysis method is not effective.Therefore,we proposed a sentiment analysis based on the Self-Attention mechanism based convolutional neural networks(CNN)and BILSTM model structure(SCBILSTM model).SCBILSTM used a bidirectional cyclic neural network to extract features from text contexts and used CNN performs local feature extraction,and then added a self-attention mechanism.On the microblog dataset captured by the web crawler,it was compared with other models to verify that the proposed model effectively improves the accuracy of text classification.