Text sentiment analysis method based on attention word vector

Abstract

In order to improve the accuracy of text sentiment classification, this paper proposes a sentiment analysis method based on attention word vector. Aiming at the problems of irregular, sparse and unclear user comments, an AT-C-GRU model combining attention mechanism, convolutional network and gated recurrent unit is proposed. The model converts the crawled user comments into word vectors and introduces the attention mechanism as input. After convolution and pooling, the sentiment classification of the comments is carried out through the GRU network, so that the key information in the user comments can be extracted more easily, and the accuracy of sentiment classification is improved. Experimental analysis shows that the accuracy of the sentiment classification method of this article on 3 data sets reaches 94.5%, 93.0% and 90.3%. This proves that the method in this paper has a good effect on sentiment classification of online review text.