Unveiling international tourist perceptions with a new human-AI collaborative approach: case of Uzbekistan

Focusing on tourism destinations in Uzbekistan, this study constructed a human-AI collaborative framework to analyze user-generated content (UGC) from Tripadvisor. By employing the DeepSeek large language model (LLM) with zero-shot learning (ZSL) and prompt engineering, combining human collaboration and validation, this study constructed a tourist perception evaluation index system and conducted sentiment-semantic analysis. First, ZSL combined with prompt engineering was used to automatically extract perception-related evaluation features that tourists focused on through the LLM. Second, by manually refining and filtering the indicators, this study constructed a hierarchical tourist perception evaluation index system comprising four primary indicators and twenty-four secondary indicators. Furthermore, based on the human-AI collaborative approach, each review was assigned sentiment scores across twenty-four evaluation dimensions. Finally, the visualized analytical results were produced to support government decision-making. This study establishes a technical framework for tourism sentiment analysis and governance, providing a practical, low-cost, and efficient tool for destination management.