Hybrid Sentiment Analysis Model for Movie Reviews

With the rapid development of natural language processing technology, BERT and CNN (Convolutional Neural Network) have achieved remarkable results in their respective fields. With the bidirectional Transformer encoder structure, BERT (Bidirectional Encoder Representations from Transformers) can effectively capture text context information and perform well in semantic understanding tasks. On the other hand, CNN is good at extracting local features through convolutional layers, which has obvious advantages when dealing with data with grid structure. However, traditional sentiment analysis methods have shortcomings in processing complex semantics and capturing local semantics, and the direct application of BERT also faces problems such as high computational cost and weak local feature capture ability. In this paper, a hybrid sentiment analysis model based on BERT and Convolutional Neural Network (CNN) is proposed to solve the above problems. By combining the advantages of BERT and CNN, the system can effectively capture the sentiment tendency of the text. Experimental results show that the accuracy of the model on IMDb movie review dataset is 91.5%, and the accuracy of the model on ChnSentiCorp dataset is 89.4%, which is significantly better than the baseline model.