A sentiment analysis method for tourist review texts based on a BERT dual-channel hybrid network

Traditional sentiment analysis methods struggle to address the challenges of colloquial expressions, semantic sparsity, and metaphorical expressions in travel reviews due to their insufficient semantic representation and feature association capabilities. This paper proposes a dual-channel model, DC-BCGM, which integrates CNN, BiGRU, and an attention mechanism. First, it uses BERT to obtain dynamic word vectors, addressing the limitations of static representations. Then, it combines BiGRU to extract global dependencies and CNN to capture local semantics, enhancing complex text representation through feature fusion from two channels. Additionally, an attention mechanism is introduced to dynamically adjust the weights of BiGRU output features, strengthening key features and suppressing noise. Finally, Softmax is applied for binary classification of the fused features from both channels. We conducted experimental evaluation on the data set of tourist attractions in Sichuan and Beijing. The accuracy of DC-BCGM model was 95.85% and 90.15%, respectively, and the accuracy of F1 score was 97.65% and 92.37%, respectively.These results demonstrate that the proposed DC-BCGM model outperforms other baseline models.