Dietary Recommendation Based on Multi-objective Optimization Integrating Nutritional Knowledge and Preference-Health Balance

We proposed a dietary recommendation method based on multi-objective optimization integrating nutritional knowledge and preference-health balance. The method captured user preference through food nutrition knowledge embedding, and employed a multi-objective optimization algorithm to balance the dietary preference of users with their nutritional health needs. Personalized recipe recommendation was performed based on the nutrition-guided food knowledge-aware network (NG-FKN), followed by nutritional food combination recommendation (NFCR). The NG-FKN extracted relevant information from the food nutrition knowledge graph, providing guidance to capture user preferences in order to achieve personalized recommendations for nutritional recipes. NFCR optimized multiple objectives in preference and health using the non-dominated sorting genetic algorithm (NSGA) based on nutritional dominance. In addition, we constructed a food-user interaction dataset, which contained 19 669 recipes, and we assessed the proposed method on this dataset. We also introduced two indicators, China healthy diet index (CHDI) and food variety score (FVS), to evaluate the nutritional value of the recommended recipes. The experimental results on the proposed dataset showed that our method was superior to current popular methods, further improving the recommendation performance of dietary nutrition.