融合营养知识和偏好-健康多目标优化的膳食推荐  

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

作　　者：金颖 闵巍庆 周鹏飞 梅舒欢 蒋树强[1,2,3] JIN Ying;MIN Weiqing;ZHOU Pengfei;MEI Shuhuan;JIANG Shuqiang(Key Laboratory of Intelligent Information Processing,Chinese Academy of Sciences,Beijing 100190,China;Institute of Computing Technology,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China;Institute of Intelligent Computing Technology,Chinese Academy of Sciences,Suzhou 215009,China)

机构地区：[1]中国科学院智能信息处理重点实验室,北京100190 [2]中国科学院计算技术研究所,北京100190 [3]中国科学院大学,北京100049 [4]中科苏州智能计算技术研究院,江苏苏州215009

出　　处：《食品科学》2025年第6期29-37,共9页Food Science

基　　金：国家杰出青年科学基金项目(62125207);国家自然科学基金面上项目(62072289);北京市自然科学基金项目(JQ24021)。

摘　　要：本实验提出一种融合营养知识和偏好-健康多目标优化的膳食推荐方法。该方法通过食品营养知识嵌入引导学习用户的偏好,并利用多目标优化算法平衡用户饮食偏好与营养健康需求。首先,基于营养引导的食品知识感知网络(nutrition-guided food knowledge-aware network,NG-FKN)进行个性化菜谱推荐,再进行营养套餐推荐(nutritional food combination recommendation,NFCR)。NG-FKN从食品营养知识图谱中提取营养信息,引导捕捉用户偏好,实现个性化菜谱推荐。NFCR结合用户饮食偏好与营养需求,采用基于营养支配的非劣排序遗传算法优化多个目标,获得营养套餐推荐列表。此外,构建食品营养知识图谱与食品-用户的交互数据集,包含19 669条菜谱数据,并在该数据集上对所提方法进行评估。同时还引入中国健康膳食指数和食物多样性评分两个指标用于评估推荐套餐的营养价值。在所提数据集上的实验结果表明,本研究方法优于目前主流的方法,可以提升膳食营养推荐的性能。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 knowledgeaware 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 19669 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.

关 键 词：膳食推荐 营养套餐 知识图谱 多目标优化 遗传算法 

分 类 号：TS201[轻工技术与工程—食品科学]