Unlocking visitor experiences in cultural heritage sites with SHAP-interpretable AI and social media sentiment analysis

This study proposes an AI-driven framework to assess visitor perceptions in historic urban areas using social media data. Unlike traditional models (e.g., XGBoost), our cascaded Aspect-Based Sentiment Analysis (ABSA) with BO-DXGBoost significantly improves prediction accuracy and interpretability. The two-stage system first classifies aspects (service quality) and then analyzes sentiment polarity and intensity, addressing class imbalance via ADASYN and RF-SMOTE. SHAP analysis visualizes feature impacts, while finer-grained interpretability is achieved by decomposing “spatial satisfaction” into 26 secondary indicators, surpassing existing coarse-grained approaches. This framework offers nuanced, data-driven insights into visitor satisfaction, supporting sustainable cultural heritage management and adaptive urban planning. By leveraging social media sentiment, it bridges the gap between big data analytics and heritage studies, enabling scalable, practical monitoring of visitor experiences in complex historic contexts.