基于BiGRU和胶囊网络的文本情感分类研究

Study on Text Sentiment Classification Based on BiGRU and Capsule Networks

为了避免池化层重要特征信息的丢失以及改善CNN和RNN无法全面提取特征的局限性,文章提出一种基于BiGRU和胶囊网络的神经网络模型—BGCapNet模型,该模型使用两个不同尺寸的BiGRU进行特征提取,实现文本长距离相互依赖的特性,胶囊网络获取更丰富的特征信息并通过胶囊预测进行情感分类。为了评估模型的有效性,在电影评论IMDB和SST-2这两个数据集上进行了实验。实验结果表明,BGCapNet模型在影评数据集上的准确率和F;值优于其他传统方法,有效提高了文本情感分类的效果。

In order to avoid the loss of important feature information in the pool layer and improve the limitation that CNN and RNN can not fully extract features,this paper proposes a neural network model based on BiGRU and capsule network—BGCapNet model.The model uses two BiGRU with different sizes for feature extraction to realize the characteristics of long-distance interdependence of text.The capsule network obtains richer feature information and classifies emotion through capsule prediction.In order to evaluate the effectiveness of the model,experiments are carried out on two data sets of film review IMDB and SST-2.The experimental results show that the accuracy and F;value of BGCapNet model in film review data set are better than other traditional methods,and the effect of text emotion classification is effectively improved.