Research On Destination Tourism Data Based on CNN-BiLSTM and STL-LSTM Methods for Data Continuity Problem

This study addresses challenges in sentiment classification accuracy and time series forecasting precision by analyzing visitor reviews and flow data for the Dunhuang Mogao Grottoes. Using a Convolutional Neural Network-Bidirectional Long Short-Term Memory (CNN-BiLSTM) model for sentiment analysis, the study demonstrated superior performance in classifying positive reviews compared to traditional machine learning and Text Convolutional Neural Network (TextCNN) models, despite slightly lower recall and accuracy for negative reviews. Time series forecasting with the Seasonal-Trend Decomposition using Loess-Long Short-Term Memory (STLLSTM) model exhibited higher accuracy for short periods (e.g., 2 days), with an average MAE of 35.71, RMSE of 59.02, and $R^{2}$ of 0.93, though accuracy decreased over longer periods (e.g., 30 days). This research provides insights into visitor sentiments and patterns, offering management and strategy recommendations to enhance visitor experiences. Future studies should refine these models to improve accuracy and efficiency.