From words to growth: Unveiling government attention to tourism from natural language processing

This paper theorizes and quantifies the government attention to tourism (GAT) using an AI-driven interdisciplinary approach to analyze government policy portfolios. By leveraging machine learning and natural language processing techniques, including textual analysis, word embeddings, and GPT-4o-based segmentation, the GAT indicator is derived from government annual reports. Within the framework of promotion tournament model and limited attention allocation theories, the study uses post-double-selection LASSO to identify key antecedents of GAT: the number of A-level scenic spots, male municipal party secretaries, and cities' economic constraints. These factors collectively shape government resource allocation in tourism policy. Validation tests confirm a positive association between GAT and actual government inputs in tourism-related domains. When governments' words align with actions, GAT can be a supplementary indicator for forecasting tourism growth. Robustness checks validate these findings, providing a reliable methodology. This study offers a comprehensive technology roadmap, guiding future tourism research with AI-driven approaches.