From Concepts to Meals: Personalized Dietary Recommendation Via Concept-Cognitive Learning and Multi-Objective Optimization

Abstract

Dietary recommendation involves trade-offs among cost, nutritional coverage, and food diversity, yet existing methods focus on single-day optimization, leaving weekly diversity, interpretability, andpersonalized adaptation insufficiently explored. This study introduces a framework integrating Formal Concept Analysis (FCA) with Pareto multi-objective optimization. Formal concepts extracted from a food-nutrient context serve as interpretable semantic units, each specifying an explicit food set and its shared nutrients. Pareto sorting preserves trade-off solutions, while user profiles and a greedy weekly strategy with novelty rewards construct seven-day meal plans. Experiments with 450 foods and 15 nutrients across six profiles demonstrate weekly food variety of 40-82 items, nutrient coverage of 84.8%-92.4%, and higher diversity and interpretability than Linear Programming, Genetic Algorithm, and Graph Neural Network baselines.