Behavioral feature-based relevance ranking for tourism-policy recommendation: A decision-support framework

This paper develops a decision support system that ranks tourism-subsidy programs for individuals according to their travel behavior. Current tourism promotion policies in Korea distribute benefits through first-come-first-served mechanisms that ignore differences in traveler preferences. We address this by constructing individual-specific, penalized behavioral features — a baseline travel-propensity term and a distance-sensitivity term — using Zero-Truncated Negative Binomial (ZT-NB) models, grouping travelers with machine-learning-based Gaussian Mixture Model (GMM) clustering, and ranking policies for each traveler through an explicit eligibility filter followed by segment similarity, geographic distance, and Bidirectional Encoder Representations from Transformers (BERT)-based semantic similarity for natural language processing. Using 2024 Korea Tourism Survey data (n=3616 multi-destination travelers), we identify seven exploratory traveler segments and assess their stability by resampling the full pipeline. In a revealed-preference leave-one-out evaluation, scoring destinations by individual-specific behavioral features improves Hit@10 over random and nearest-neighbor baselines, but does not exceed a coarse region-match heuristic or a homogeneous-data artifact control, and remains below a simple pooled distance coefficient; the pooled (sample-level) distance effect and the single largest segment are the components that survive these controls. We therefore frame the artificial-intelligence contribution as exploratory behavioral segmentation and relevance ranking rather than demonstrated individual-level personalization, and the engineering application as a decision-support prototype rather than a validated mechanism for increasing travel uptake.