Integrating topic clustering and sentiment analysis for visitor perception evaluation in historic blocks: insights from the Dadou Road case

Historic blocks play a crucial role in sustainable urban and rural development by preserving cultural heritage while adapting to contemporary needs. However, visitor perspectives – particularly regarding spatial perceptions and satisfaction – remain underexplored, and few studies integrate social media data with advanced natural language processing techniques. This study applies BERTopic for thematic clustering of social media posts to identify key aspects of visitor experience, followed by a fine-tuned BERT sentiment analysis model that classifies comments into three categories – positive, neutral, and negative. To enhance validity, on-site interviews were conducted for qualitative verification, enabling comparison between online sentiment patterns and in-person perceptions. The insights derived from the analysis are translated into Importance-Performance Analysis (IPA) quadrants, furnishing actionable guidance for heritage managers in both urban and rural contexts to optimize conservation strategies and elevate visitor experiences. This research underscores the potential of deep learning-based text analytics for the stewardship of the built environment heritage, thereby proposing a replicable framework for evidence-based and contextually attuned decision-making.