Quantifying spatial perception in the adaptive reuse of cultural-creative industrial heritage using user-generated text in Chongqing

Adaptive reuse of industrial heritage requires operational-stage evidence on public spatial perception, yet scalable and reproducible assessment remains limited. We developed a user-generated content (UGC)-based text-mining framework and applied it to eight cultural-creative industrial heritage sites in central Chongqing, China. Fourteen indicators covering industrial characteristics, functional ecology, public attraction, cultural communication and spatial resilience were quantified using word frequency, SnowNLP sentiment analysis and semantic association, then integrated using the entropy weight method (EWM), with equal weighting and multiple robustness checks. In a 400-instance human audit, the two annotators agreed on 80.25% of labels. SnowNLP agreed with 73.52% of the 321 strict-consensus labels, although chance-corrected and class-balanced metrics indicated limited three-class validity. Across 1000 within-case bootstrap replicates, the probabilities of entering the top three were 90.3% for JZS, 86.8% for ZBC and 81.5% for JSP. Exact ordering varied with weighting, sampling, case composition and model specification. The framework therefore offers a reproducible and quantitative public-perception perspective for operational-stage diagnosis of cultural-creative industrial heritage reuse, complementing field-based assessment while making sample dependence explicit.