Global and regional inequalities in dairy recommendations: a natural language processing analysis of food-based dietary guidelines across income groups.

Background: Food-based dietary guidelines (FBDGs) are key public health instruments aimed at promoting healthy dietary patterns. However, dairy-products recommendations vary substantially across countries, reflecting not only scientific evidence but also socioeconomic conditions, institutional capacity, and food system characteristics. The extent to which these differences are structured linguistically across income levels has not been systematically quantified. The present study aims to explore the use of advanced NLP techniques to characterize semantic differences and similarities in dairy-products dietary messages within FBDGs from countries with different levels of socioeconomic development. Methods: We conducted a comparative analysis of dairy-products recommendations extracted from national FBDGs officially recognised by the Food and Agriculture Organization, covering 98 countries. Using advanced natural language processing techniques, including lexical frequency analysis, co-occurrence networks, and latent topic modelling, we examined semantic patterns in recommendation statements and justificatory texts. Countries were stratified according to World Bank income group classifications. Findings: Across all income groups, "milk" emerged as the central lexical anchor of dairy recommendations. However, high-income countries demonstrated greater lexical diversity and semantic complexity, incorporating differentiated references to product types, fat content, and fermentation (e.g., yogurt, cheese, low-fat). In contrast, low- and lower-middle-income countries presented more general and nutritionally basic messaging, primarily focused on consumption adequacy and child nutrition. Justification texts consistently contained higher nutrient-related terminology density than recommendation statements. Interpretation: Dairy-products dietary messaging in national FBDGs shows consistent descriptive differences across income groups. These semantic disparities likely reflect contextual differences in institutional capacity, epidemiological priorities, and food system infrastructure. The findings may be particularly relevant for the Americas, where high-, upper-middle-, and lower-middle-income countries coexist within a region undergoing rapid nutrition transitions and facing the persistent triple burden of malnutrition. In this context, PAHO/WHO may play an important role in supporting greater harmonisation of dietary guidance across diverse socioeconomic settings. NLP-based approaches offer scalable tools for monitoring global nutrition policy discourse and supporting evidence-informed policy development. Funding: Research supported by the Vice-Presidency of Research and Doctoral Studies of Universidad San Sebastián, Grant USS-FIN-26-APCS-01; Institutional collaboration provided by the Scientific Committee of Dairy Products of the Chilean Dairy Consortium (Consorcio Lechero) through the "Gracias a la Leche" program.