Mapping the Landscape of Nutrition Recommendation Systems: Evidence from Peer-Reviewed Research, Apps, and Patents

Nutrition recommendation systems (NRSs), which integrate user data with nutritional knowledge to generate individualized advice, have emerged as promising digital tools. However, challenges remain in design, implementation, and clinical applicability. We conducted this review to map the development, characteristics, and technological aspects of NRS, and to identify existing gaps in application and evaluation. To provide a comprehensive overview, we included both peer-reviewed literature and non-peer-reviewed sources, thereby reflecting the breadth of existing innovations beyond academic research. We systematically searched bibliographic databases, patent repositories, and software stores. From all identified NRSs, we extracted publication year, topic, interface users, input variables, system-generated output, and target population. For NRSs reported in peer-reviewed studies, we further collected detailed data on author affiliations, system characteristics, evaluation strategies, artificial intelligence techniques, and recommendation algorithms. Results were synthesized and presented in visual formats. The protocol for this study was registered on the Open Science Framework (doi: 10.17605/OSF.IO/VF7NB). A total of 878 NRSs were identified, with 43.4% released after 2022. Systems mainly targeted general people or the population with overweight and were based on general information (eg, dietary habits, exercise types). Among all of the 49 NRSs published in academic studies, only 4 involved nutritionists, and nearly half relied on public surveys without documented data procedures or quality control. Most NRSs focused on nutrition advice (53.1%) as a primary output. Evaluation relied primarily on internal test sets (22.4%). Accuracy (34.7%) was primarily metric. Convolutional Neural Networks (14.3%) and Random Forests (14.3%) were the top smart techniques; most models were non-self-updating. Content-based filtering (30.6%) dominated recommendation algorithms, with the latest proposed algorithm dating to 2014. Current nutrition recommendation systems lack personalization, standardized evaluation, and nutrition expert involvement. Most systems rely on general data and outdated algorithms, limiting their clinical relevance and applicability. Enhancing individualization, ensuring data transparency, and fostering interdisciplinary collaboration are critical to improving the effectiveness and reliability of future systems.