From preferences to narratives: regional preference and semantic structures in Weibo discussions of Beijing’s 5A attractions

Abstract Digital traces from social media increasingly shape how tourist preferences are observed and interpreted, yet attraction-related attention is often treated as a single aggregate indicator, with limited consideration of where this attention comes from and how attractions are discursively framed. Focusing on eight 5A attractions in Beijing, this study uses Sina Weibo posts from 2019 to 2023 to construct a province–attraction–time database and derive two complementary measures: a total preference intensity $$F$$ , capturing emotion-weighted attention, and a preference index $$L$$ , capturing excess or deficit preference relative to an independence benchmark. A lexicon-based sentiment pipeline is used to obtain post-level scores, and temporal patterns are analysed at monthly and seasonal scales. Spatial distributions of $$F$$ and $$L$$ are explored through mapping and clustering, while structural drivers are examined using a gravity-type Poisson model for F and a geographical XGBoost model for $$L$$ interpreted with SHAP values and local importance maps. The results show persistent coastal and Sichuan-centred hotspots of preference, gradual diffusion towards parts of central and south-western China, and mostly positive but event-sensitive sentiment trajectories. Population, income and tourism service supply strongly increase total preference intensity, whereas rich local consumption and accommodation environments are associated with weaker outbound preference; distance exerts only a weak damping effect. Semantic analysis reveals stable clusters of attraction-related discourse and distinct emotional profiles within them. Together, the framework demonstrates how social-media data can be used to characterise uneven regional engagement and divergent semantic constructions of iconic attractions in contemporary China.