Research On Diet Recommendation Algorithm Based On Knowledge Graph And Collaborative Filtering

In recent years,with the advancement of medical technology and the improvement of people’s living standards,the average life span of human beings has been greatly extended.However,many people have long-term unhealthy eating habits,the incidence of some diseases,such as diabetes,cardiovascular disease,peptic ulcer and gastroenteritis,has greatly increased.In order to help people keep healthy eating habits and reduce the incidence of these diseases,we propose a healthy recipe recommendation algorithm based on knowledge graphs and collaborative filtering.The main contributions of this work are summarized as follows:1.We construct a knowledge graph for healthy recipe.Firstly,the experimental data in the diet field is preprocessed according to the data source and data type.In this paper,the schema layer of knowledge graph is defined by data analysis and nutrition experts’ suggestions;The entities in textual data are extracted by Dic+CTR model which is proposed in this paper.the relationship between entities is classified using the BiLSTM model and the weighted dictionary matching method;Since same entity may be different in different data sources,we aligns the entities obtained;Finally the knowledge graph is stored in the OWL ontology language,and the construction of the knowledge graph in the diet field is completed.2.Vector representation of Recipe based on TransHR model.By embedding the semantic information between the recipes into the low-dimensional vector space,the semantic similarity between the recipes can be calculated.Since traditional knowledge representation models cannot effectively deal with the complex relationships between entities in the knowledge graph for healthy recipe,we select an improved hyper-relational model for knowledge graph embedding,which is called TransHR.After link prediction experiments and triple classification experiments,the experimental results show that the model has a good performance on the recipe knowledge graph.3.We proposed a recipe recommendation algorithm joint knowledge graph and collaborative filtering.Since the traditional collaborative filtering-based recipe recommendation algorithm only based on the user-item scoring matrix and does not uses the semantic information of the item,users’ ratings on recipes are not good indicators for healthiness,as many recipes with high ratings are pretty unhealthy.We introduce the semantic information between the recipes as an important recommendation basis by constructing a knowledge graph.By representing the entity and relationship of the recipes in two different low-dimensional continuous vector spaces,the semantic similarity between the recipes is calculated,and the semantic similarity and the collaborative filtering similarity are finally combined for recommendation.Compared with the traditional recipe recommendation algorithm,our proposed algorithm fully utilizes semantic information,alleviates the data sparsity and cold start problems.The experimental results on the diet data set show that the algorithm in this paper has a significant effect on the diet recommendation task,and has significantly improved the recall rate and AUC value.