Regenerative hospitality model using tourist municipalities using web scraping and unsupervised machine learning

Based on the research question: How can the regenerative hospitality actions undertaken by municipalities listed on the tourism map be classified? This study proposes a classification model for regenerative hospitality in Brazilian tourist municipalities. To achieve this aim, the study adopted a municipal-level adaptation of the regenerative dimensions proposed by Inversini et al. (2023), followed by the adapted model. Data were collected using natural language processing techniques, supported by a Python-based web‑scraping system that extracted text from municipal webpages listed on the National Confederation of Municipalities tourism map. The system gathered data, computed cosine similarity, vectorised the text, and compared the results against entries stored in a MySQL database, according to the dimensions of the regenerative hospitality model. The system obtained approximately 14,000 records and aggregated them by municipality, federative unit, and dimension. The dataset was then subjected to exploratory factor analysis using the principal components method with varimax rotation, which yielded two dimensions—socio‑environmental regeneration (X‑axis) and regenerative co‑creation (Y‑axis)—that explained 68% of the variance. K-means clustering was then applied, and the Elbow method indicated a three‑cluster solution: Cluster 1—Connectivity and Ideation, Cluster 2—Expanded Regenerative Core, and Cluster 3—Initial Base (Regeneration). The proposed framework offers three contributions. Theoretically, it tests the boundary conditions under which a firm-level regenerative hospitality model can be adapted to a territorial governance unit. Methodologically, it demonstrates a reproducible, partially validated NLP pipeline for classifying public-sector managers. Practically, it gives public managers and researchers a tool to assess, compare, and prioritise regenerative action across municipalities.