基于多模态的冷启动饮食推荐算法研究与实现

RESEARCH AND IMPLEMENTATION OF COLD START DIET RECOMMENDATION ALGORITHM BASED ON MULTI-MODAL

为帮助用户在海量食谱中找到符合个人饮食偏好食谱,提出一种基于多模态的饮食偏好推荐冷启动算法。该算法分为线上和线下两个阶段。线下阶段,通过表示学习获得食谱各维度信息的语义向量。线上阶段,基于迭代、图相似度传播的思路,通过少量食谱的偏好反馈获得用户对全部食谱的偏好度并推荐。实验结果表明,基于多模态的冷启动饮食推荐算法在实验平台上的准确率达到实际工程应用水平。

In order to help users find recipes in line with their personal dietary preferences from a large number of recipes,a cold start algorithm for dietary preference recommendation based on multi-modal is proposed.The algorithm was divided into online and offline phases.In the offline stage,the semantic vector of each dimension of recipe information was obtained through representation learning.In the online stage,based on the idea of iteration and graph similarity propagation,the user's preference for all recipes was obtained and recommended through a small amount of preference feedback of recipes.The experimental results show that the accuracy of the cold start diet recommendation algorithm based on multi-modal achieves the level of practical engineering application.