Integrating Tourist Satisfaction into Recreational-Service Diagnosis and Visual Optimization Simulation in National Forest Parks Using vLLMs

Demand for nature-based recreation has increased the need for recreational-service planning in National Forest Parks (NFPs). Existing studies evaluate recreational services, but translating findings into actionable planning remains challenging. This study developed a vLLM-assisted framework linking indicator prioritization, image-based diagnosis, optimization simulation, and model-based reassessment. Questionnaire data comprised 238 responses from eight NFPs in Hangzhou, with 23 predictors measured predominantly at the park level. SHAP analysis ranked the recreational-service dimensions by predictive importance: facility (25.6%), activity (21.6%), tourist use experience (19.7%), information (17.5%), and accessibility (15.6%). Their weights guided image-based assessment of 73 NFPs across the Yangtze River Delta. Of 353,608 tourist-generated images collected, 96,876 passed screening. Human-reference evaluation showed correlations between model and human dimension scores of 0.815–0.898. Image diagnosis highlighted visible problems in signage, resting and public-service facilities, environmental interpretation, activity-space organization, and path legibility. Optimization simulation increased the mean model-assessed composite score of 365 selected low-scoring scenes from 6.26 to 7.32 (16.93%), with higher park-grouped means in 71 of 73 NFPs. The framework provides a scalable pathway for translating satisfaction-informed priorities into scene-level recreational-service diagnosis and visual planning support.