基于多维度特征的饮食健康推荐

MULTI-DIMENSIONAL RECOMMENDATION ON HEALTHY DIET

为帮助人们判断饮食是否合理或营养是否均衡,进而改善饮食结构,合理化不同营养元素的摄入量,提出一种基于用户画像、饮食记录和即时情境多维度特征的健康饮食推荐算法。基于用户画像及用户的静态特征,制定不同人群、疾病的膳食指南,得到满足该用户身体状态的推荐列表;基于饮食记录,采用聚类方法和改进的距离计算方法,挖掘用户的营养结构和口味偏好,能够合理有效地进行饮食推荐;基于即时情境,通过外部特征对前两种算法推荐列表进行重排序,产生符合当前用户状态的菜品。实验结果表明,多维度特征推荐算法在实验平台上准确率达到实际工程应用的水平。

In order to help people judge whether the diet is reasonable or nutritionally balanced,and then improve the diet structure and rationalize the intake of different nutrients,we proposed a healthy diet recommendation algorithm based on user portraits,dietary records and multi-dimensional features of instant situations. Based on the portrait and the static characteristics of the users,we formulated the dietary guidelines for different populations and diseases,and obtained the recommended list to meet the user’s physical condition. Based on the dietary records,we used the clustering method and the improved distance calculation method to mine the user’s nutritional structure and taste preference,which could reasonably and effectively recommend diet. According to the instant situation,the first two algorithms recommendation lists were reordered by external features to produce dishes that conform to the current user’s taste. The experimental results show that the accuracy of multi-dimensional feature recommendation algorithm reaches the level of practical engineering application.