Big Data Applications and Intelligent Decision Support Framework for Study Tourism Market Demand Forecasting

Abstract This paper addresses three critical bottlenecks in study tourism market demand forecasting: data heterogeneity, time series non-stationarity, and decision-making lag, proposing a big data-driven intelligent decision support framework termed BDIDS (Big Data Intelligent Decision Support). The framework comprises a three-tier architecture consisting of a data fusion layer, a predictive computation layer, and a decision output layer. The data layer integrates multi-modal sources including travel platforms, social media, and policy documents through standardized feature engineering pipelines fitted exclusively on training-period observations to prevent leakage. The predictive layer introduces a cross-modal attention fusion mechanism, coordinating BERT semantic embeddings with time series statistical features to drive an improved Transformer-LSTM hybrid model. The decision layer embeds demand elasticity analysis and supply-demand matching optimization modules. Using a provincial-monthly panel dataset from four East China provinces/municipalities spanning 2018–2024, 336 total province-month cells are constructed; after excluding 24 cells corresponding to complete operational suspension during COVID-19 control periods, 312 valid province-month observations remain, partitioned into 216 training, 48 validation, and 48 test observations (216 + 48+48 = 312), with the full panel index provided as supplementary material for independent verification. On the 48-observation 2024 test set, BDIDS achieves RMSE reductions versus ARIMA and single LSTM under a pre-specified Diebold–Mariano protocol (MSE loss, Newey–West HAC lag = 4, Bonferroni-corrected across seven comparisons); results are conditional on training-period specification and constitute within-sample evidence. Accuracy metrics in ten thousand person-trips support industry resource allocation decisions.