Abstract While traditional visitor perception studies rely heavily on questionnaire surveys with limited samples, there is a lack of data-dr
Abstract While traditional visitor perception studies rely heavily on questionnaire surveys with limited samples, there is a lack of data-driven approaches that can comprehensively capture and analyze large-scale visitor feedback. This study employs machine learning methods, particularly the Latent Dirichlet Allocation topic model, to analyze large-scale User Generated Content data collected from Xiecheng.com, one of China's largest online review platforms. Based on the case study of Wuhan Garden Expo Park, this research constructing a multi-dimensional evaluation framework for visitor perceptions in suburban parks. Analysis of 3,099 valid reviews identified four main dimensions of visitor perception: service quality and overall evaluation, natural landscape, recreational activities and entertainment, and cultural and educational experiences. The study quantitatively assessed attention and satisfaction levels across these dimensions. Service quality and overall evaluation received the highest attention (36.1%) and high satisfaction (19.19), while natural landscape showed moderate attention (26.8%) and satisfaction (17.74). Recreational activities and entertainment exhibited a “high attention-low satisfaction” pattern, revealing potential improvement. Cultural and educational experiences presented a “low attention-high satisfaction” characteristic, suggesting untapped advantages. These findings validate the multi-dimensional nature of visitor perceptions and highlight suburban parks' unique attributes. The results provide data-driven insights for park management, emphasizing balanced resource allocation and visitor expectation management. This study demonstrates the value of machine learning in tourism research, offering a new methodological perspective for visitor perception studies.
