Mixed network sentiment analysis combining sentiment features and multiple attention

In order to solve the problem of low attention of some emotional words in sentiment analysis tasks and difficulty in capturing long-distance dependence between sentences, this paper proposes a mixed sentiment analysis network (DB-BGA-CNN) that integrates multiple attention mechanisms of sentiment words. First, a more targeted emotional dictionary is obtained by expanding the dictionary, and an emotional word selection segmentation algorithm (DSS) is designed. Secondly, use Bert to encode the word vector of the sentiment words and phrases selected from the sentence and sentiment dictionary respectively to obtain the deep semantic features of the text and perform fusion. Then, use multiple attention mechanisms to realize the enhancement of sentiment analysis capabilities, and discuss which network effect is better; finally, the output vectors of each network are merged, and the activation-pooling layer is used to avoid the occurrence of overfitting. Compared with multiple existing models, the proposed model shows better performance, and the accuracy of the optimal model reaches 95.80%.