Deep learning-driven evaluation of public Perception toward China’s railway parks using web-based textual data

Railway parks have become significant public spaces within urban transformation and sustainable development, offering unique infrastructure and recreational value. This study investigates the perceived characteristics of China’s railway parks to guide improvements in visitor experiences. Web scraping techniques collected user reviews, analyzed with the BERTopic model to identify seven perception topics. Multi-label classification with Baidu’s deep learning model enabled text categorization and sentiment polarity analysis. Results reveal that railway heritage resources and recreational activities are the dominant perception topics. High-quality heritage and natural landscapes drive positive emotional responses, while issues related to accessibility, usability, and attractiveness elicit negative experiences. Significant differences in public perceptions were observed across park types, preservation-focused parks and those combining heritage with natural landscapes received more favorable evaluations. These insights offer a basis for enhancing railway park design and development through perception-driven evaluations.