AI-Driven Framework for Assessing Visitor Perceptions in Historic Urban Areas Using Social Media

As urbanization progresses, preserving and adapting historical districts for sustainable development is crucial. These areas embody significant cultural heritage and contribute to economic, social, and cultural sustainability. However, research on visitor perceptions, particularly spatial satisfaction, is limited, especially in fine-grained analyses using social media data. This study introduces a framework for evaluating visitor perceptions using Aspect-Based Sentiment Analysis (ABSA) enhanced by a BO-DXGBoost model, a cascaded system combining two XGBoost models fine-tuned through Bayesian Optimization (BO). The first model identifies aspect categories, while the second analyzes sentiment polarity and intensity. Class imbalance is addressed using ADASYN and RF-SMOTE, and SHAP analysis visualizes feature influences on sentiment predictions. This framework provides quantitative insights into visitor perceptions of historical districts and offers a robust approach for sustainable heritage management through the integration of social media data.