A Sentiment Classification Model Based on Deep Learning

Sentiment classification is an interesting and crucial research topic of opinion mining and sentiment analysis, which is used to obtain sentiment types from the text documents of numerous sources. Sentiment classification has the problems of insufficient semantic feature extraction and ignoring context information. This paper proposes a MSCNN-BiGRU model for text sentiment classification by utilizing multi-scale convolution kernels to extract rich semantic features and BiGRU to extract features containing text context information. The experimental results on the Chinese Weibo text dataset and the English e-commerce reviews dataset show that the classification accuracy of this model is 3.41% better than CNN and 3.59% better than RNN. The model proposed in this paper is not only suitable for text sentiment classification, but also helpful for expression classification research with time series information.