Tourism Consumer Sentiment Analysis Using a Multi-layer Memory Network Combining Temporal Convolutional and BiLSTM Architectures

Traditional sentiment analysis methods for tourism review texts often suffer from a strong dependence on vocabulary quality and insufficient ability to capture contextual temporal features, making it difficult to accurately identify sentiment tendencies in complex contexts. To address this issue, this study proposes a multi-layer memory network algorithm that integrates temporal convolutional networks (TCN) and bidirectional long short-term memory (BiLSTM) to improve the depth of feature extraction and the accuracy of emotion classification. A dataset of 30,000 manually labeled tourism review texts was constructed, with sentiment labels categorized into positive, neutral, and negative. Experimental results showed that the proposed model achieved an accuracy of 95.78%, an F1 score of 95.98%, and a training time of only 40.18 seconds in sentiment classification tasks. Ablation studies demonstrated that the model exhibited clear advantages in structural integration and regularization mechanisms, with stronger stability. Sentiment orientation analysis showed that positive reviews account for 74.8% on average, and the model effectively identified multi-emotional expressions, indicating strong practicality and generalization ability. In summary, the proposed model achieves excellent performance in both feature extraction and sentiment classification, and achieves high generalization and training efficiency. Its application to tourism consumer review analysis can provide reliable technical support for the precise management of tourist attractions and user feedback analysis, while also offering new insights for the advancement of sentiment analysis methods.