Dynamic Word Vector Based Sentiment Analysis for Microblog Short Text
Zhang Peiliang, Mijit Ablimit, Hankiz Yilahun, Askar Hamdulla · 2022
Aiming at the disadvantage of static word vector model to the problem of polysemy of words, a model based on dynamic word vector and two-way GRU neural network combined with attention mechanism is proposed to recognize emotion of microblog short text comments, which is helpful to analyze and monitor social public opinions. First, the word vector is obtained through ERNIE pre-training model, and then transmitted to the bidirectional GRU neural network. Then, the keywords are weighted in combination with the Attention mechanism. Finally, the emotion is classified through the softmax layer. Experimental results show that the classification accuracy of the proposed model on the weibo_senti_100k data set is 96.51%.