<![CDATA[This study implemented a personalized healthy food menu recommendation system using content-based filtering and cosine similarity. The system translated user characteristics, dietary goals, preferences, allergies, and health-related restrictions into menu attributes. The pilot dataset contained 10 healthy menus represented by 15 binary attributes. Recommendation performance was evaluated offline using 10 controlled user-profile scenarios with realistic age, sex, weight, height, and activity values. Basal metabolic rate and total daily energy expenditure were calculated with the Mifflin-St Jeor equation. After allergen filtering, 94 user-menu candidate pairs were assessed. Reference relevance labels were determined before ranking based on meal-category compatibility, absence of allergen conflicts, and fulfillment of at least two target attributes. At a cut-off of three recommendations, the system achieved macro Precision@3 of 0.7000, Recall@3 of 0.9167, F1@3 of 0.7481, nDCG@3 of 0.8948, and Hit Rate@3 of 1.0000. All 10 functional test cases passed, and no allergen-conflicting menu appeared in the final top-three results. The findings indicated that the system produced relevant and traceable recommendations, although the small pilot dataset limited generalizability.]]>
Copyrights © 2026
