Text Sentiment Analysis based on BERT and Convolutional Neural Networks

The rapid development of the network has accelerated the speed of information circulation. Analyzing the emotional tendency contained in the network text is very helpful to tap the needs of users. However, most of the existing sentiment classification models rely on manually labeled text features, resulting in insufficient mining of deep semantic features hidden in the text, and it is difficult to improve the classification performance significantly. This paper presents a text sentiment classification model combining BERT and convolutional neural networks (CNN). The model uses BERT to complete the word embedding of the text, and then uses CNN to learn the deep semantic information about the text, so as mine the emotional tendency towards the text. Through verification on the large movie review dataset, BERT-CNN model can achieve an accuracy of 86.67%, which is significantly better than traditional classification method of textCNN. The results show that the method has good performance in this field.