Healthy Recipe Recommendation using Nutrition and Ratings Models
Healthy Recipe Recommendation using Nutrition and Ratings Models
复制标题
使用营养和评级模型推荐健康食谱
DOI：
--
复制
发表时间：
2019
期刊：
影响因子：
--
通讯作者：
Nicholas Lai
中科院分区：
文献类型：
--
作者：
Y. S. Tang;An Zheng;Nicholas Lai
我们提出了一个由三部分组成的算法，根据一个人的食物偏好提供更健康，更美味的食谱替代品。第一部分包括使用线性回归开发一个营养模型，以了解给定成分的食谱的整体营养含量。第二部分包括使用图形神经网络（GNN）对成分配方和配方用户二分图进行用户评分建模。我们将这两个模型联合收割机结合起来，使我们能够根据用户的喜好来评估新食谱的健康性和美味性。我们表明，我们对评级模型的GNN方法实现了强大的性能，并提供了令人信服的定性结果，我们推荐的食谱。
We propose a three-part algorithm to provide healthier and tastier recipe alternatives given a person’s food preferences. The first part consists of developing a nutritional model using linear regression to understand the overall nutritional content of a recipe given its ingredients. The second part consists of modelling a user’s rating scores using a graph neural network (GNN) on ingredient-recipe and recipe-user bipartite graphs. We combine these two models to enable us to evaluate the healthiness and tastiness of novel recipes according to users’ preferences. We show that our GNN approach towards the ratings model achieves strong performance and provide compelling qualitative results of our recommended recipes.