Abstract: In view of the traditional collaborative filtering-based recipe recommendation algorithm that only uses the user-item score matrix and does not consider the semantic information of the item itself resulting in low recommendation accuracy, this paper introduces the semantic information between recipes as an important recommendation basis by constructing a knowledge graph, and proposes a personalized diet recommendation algorithm based on knowledge graph embedding and collaborative filtering. By representing the recipe entity and relationship in two different low-dimensional continuous vector spaces, the semantic similarity between the dishes is calculated, and the semantic similarity is incorporated into the collaborative filtering recommendation for recommendation. The method in this paper alleviates the problems of data sparsity and cold start by strengthening the use of hidden information between dishes, and makes the recommendation result more reasonable. Experiments on the dataset show that the method has a significant effect on recipe recommendation, and it has a significant improvement in recall and AUC.

Key words: recipe recommendation, collaborative filtering, knowledge graph, representation learning, semantic similarity