Tourism review texts reflect tourists’ feelings on scenic spots, hotels, transportation, catering and services, and also provide a calculation basis for satisfaction prediction. This paper integrates vocabulary normalization, contextual semantic coding, polarity recognition, aspect extraction and rating regression to construct a Natural Language Processing (NLP) based travel review sentiment analysis and satisfaction prediction network. This study selected 48,620 English travel reviews to complete text cleaning, label mapping and context embedding. The model uses BERT to extract semantic representation, and uses BiLSTM to depict the before and after semantics. Combined with attention weighting, multi-dimensional semantic fusion and consistency calibration, the model outputs three types of emotion labels and five levels of satisfaction scores. The experiments were completed in Python 3.10 environment, and the training, validation and test sets were divided into 7:1:2. The results show that the accuracy rate of the model is 94.18%, the precision rate is 93.47%, the recall rate is 92.86%, the F1-score is 93.15%, the AUC is 0.962, and the MAE is 0.218, which is better than the baseline models such as TF-IDF-SVM, CNN, BiLSTM and BERT. It can provide data support for tourism review understanding and tourist experience analysis.
