Three distinct profiles of barriers to physical activity during pregnancy among Chinese women: a latent profile analysis.

Growing evidence linking diet, obesity, and non-communicable diseases (NCDs) has increased the emphasis on adopting healthier lifestyles. In particular, analyses link dietary patterns to adolescent overweight and obesity (), while global reviews characterize its epidemiology and pathogenesis (). The World Health Organization (WHO) estimates that at least 2.8 million deaths annually are attributable to overweight or obesity and that ~2.3% of global Disability-Adjusted Life Years (DALYs) are linked to excess weight (). NCDs (e.g., cardiovascular disease, diabetes) account for 71% of deaths, many of which are preventable by addressing shared risk factors such as unhealthy diets (). At the same time, the shortcomings of a “one-size-fits-all” approach to nutrition have become evident, as individuals vary widely in their physiology, preferences, culture, health status, and behavior. These challenges have spurred the development of personalized nutrition, delivered by clinicians and increasingly supported by digital tools. Evidence from a large personalized nutrition intervention delivered online shows that personalized nutrition advice is effective (). Building on traditional approaches and the rapid advancement of Artificial Intelligence (AI) and Machine Learning (ML) in food and nutrition (), recommender systems can now tailor diets using anthropometric data, dietary choices, health conditions, personal preferences, real-time behaviors (e.g., dietary intake, adherence), and sensor streams (e.g., physical activity, continuous glucose monitoring) (–). Broadly, food and nutrition recommenders fall into two families: and (). draw on combinatorial optimization (e.g., knapsack, integer/linear programming), content-based filtering, collaborative filtering, and hybrids. Combinatorial analysis supports meal planning by balancing nutrition, cost, and user preferences and by sequencing meals to respect constraints (, ). In particular, optimization methods such as the knapsack algorithm, integer programming, and constraint satisfaction are used to select optimal sets of foods while enforcing dietary diversity, food-group rules, and user restrictions (). Although such methods can yield mathematically optimal plans, they often struggle to accommodate rich user goals, detailed rules, and diversity at scale (, ). Other traditional recommenders rely on content-based, collaborative, or hybrid techniques (, ). Content-based approaches match items to user- or item-side attributes (, ), collaborative methods infer preferences from similar users (), and hybrid approaches combine both to improve accuracy and coverage ().