Human–AI Collaborative Recommenders for On-Site Cultural Tourism

Cultural venues require scalable yet individualized experiences, but existing tourism HCI studies rarely quantify how algorithmic personalization affects real-world visitor behavior and psychometric outcomes. The authors designed a human–AI collaborative recommender that fuses visitor psychographics (Big-Five, novelty-seeking), real-time indoor location, and contextual constraints (crowd, weather) to generate adaptive itineraries delivered through a WeChat mini-program. A 14-day mixed-methods field deployment compared the system with default routes at the Palace Museum (n = 312) and Universal Studios Beijing (n = 298). Multilevel modeling revealed that the algorithmic condition increased spatial entropy by 27%, dwell time at high-value exhibits by 34%, and Flow Short Scale scores by 0.82 SD (all p < .001), while Net Promoter Score increased by 19 points. Reflexive interviews showed that explanatory AI nudges moderated trust and compliance, preserving visitor agency.