Unlocking Visitor Experiences in Cultural Heritage Sites with SHAP-Interpretable AI and Social Media Sentiment Analysis

As urbanization continues to accelerate, the preservation and adaptation of historic urban areas for sustainable development has become an urgent priority. These areas are rich in cultural heritage. They play a key role in promoting economic, social, and cultural sustainability. Despite this, research focused on understanding visitor perceptions, particularly in relation to spatial satisfaction, remains scarce. This is especially true when fine-grained analyses use social media data. Existing methods fail to effectively capture the nuanced perceptions of visitors in these complex environments. This study introduces a novel AI-driven framework for assessing visitor perceptions. The framework uses Aspect-Based Sentiment Analysis (ABSA) with a BO-DXGBoost model. This is a cascaded system that integrates two XGBoost models fine-tuned using Bayesian Optimization (BO). The first model categorizes aspects. The second model analyzes sentiment polarity and intensity. Class imbalance is tackled using ADASYN and RF-SMOTE. SHAP analysis is used to visualize how features influence sentiment predictions. This framework offers quantitative insights into visitor perceptions of historic urban areas. It provides a scalable and practical tool for sustainable heritage management by using social media data.