Multi-feature Microblog Sentiment Analysis based on BERT-AttBiGRU model

Abstract

In recent years, there has been an increasing amount of research on Weibo sentiment analysis techniques, but less attention has been paid to emoji. However, emoji are closely related to Weibo sentiment. So to judge the microblog sentiment tendency more accurately, in this paper, we select Weibo comments that contain a large number of emoji and propose a neural network classification model that combines emoji. The model first obtains word vectors containing contextual semantic information through the BERT pre-training model, then extracts deep-level feature information by using bi-directional gated recurrent network (BiGRU), and then puts the emoji vector and text vector into the Attention mechanism, and assigns weights to the extracted feature information to highlight the important information. Finally, the Softmax function is used to classify the microblog sentiment.The experimental results prove that the accuracy of the model reaches 97.65%, which effectively improves the accuracy of microblog sentiment classification.