Research on Sentiment Analysis of User Reviews Based on BERT and Parallel Neural Networks

In recent years, with the increasing prominence of environmental and energy issues, people’s demand for environmentally friendly transportation is gradually increasing, providing a broader market space for new energy vehicles. In this paper, we chose Auto Home as the research object to explore the users' emotional orientation towards new energy vehicles. In order to solve the problem of poor extraction of text features by traditional methods, a hybrid BERT-CNN+BiLSTM parallel network model incorporating an attention mechanism is designed. User review data was collected and pre-processed, we labeled the reviews with emotion according to the different star ratings. Second, the word cloud was used to analyze the processed data to visualize the high-frequency keywords. Once again, the processed data is imported into the model for text sentiment discrimination. Among them, BERT captures the contextual relationship between words and provides rich semantic information, CNN extracts key local features in text through multiple convolutional kernels, and BiLSTM focuses on capturing full text feature information. Based on this, the attention mechanism is introduced, and the fused output of CNN and BiLSTM is used as the input of the attention mechanism, and the final sentiment classification results are obtained. Compared with the traditional model, it has higher accuracy and precision and performs well on this dataset.