Design And Implementation Of Recipe Recommendation System Based On Diet Knowledge Graph

With the improving living standards,diet is playing an increasingly important role in people's family life.Meanwhile,with the rapid development of the Internet and mobile technology,it has become a daily habit for people to obtain food-related information from the Internet.The fact that the amount of network information data has grown exponentially has caused people to face serious information overload problem.Personalized recommendation is an effective way to solve this problem.Though recipe is an important carrier of dietary information,there is few personalized recommendation research in recipe field.The recipes from the popular food website are mainly recommended by the factors such as click volume and popularity without considering the user's personalized preference.In addition,with the development of knowledge graph technology,the research on combining personalized recommendation with knowledge graph is also increasing.The profound semantic information in knowledge graph helps to improve the effect of traditional personalized recommendation method.In terms of personalized recipe recommendation,the use of domain knowledge graph containing information on recipes,ingredients,nutrition and health is also beneficial to enhance the effectiveness of recommendation.In order to meet the requirements of recommending recipes for users in daily life,to solve the problem of lack of personalized recipes recommendation system,and to make full use of the advantages of knowledge graph in improving personalized recommendation,a personalized recommendation model based on knowledge graph is proposed in this thesis.The main tasks are as follows:(1)The user-recipe preference dataset is constructed by using the recipes crawled from the Meishijie website,which are collected,published and created by users.Then,the alignment of the recipe entities in the diet knowledge graph with the recipes in the user-recipe preference dataset is implemented.(2)A personalized recommendation model combining knowledge graph is proposed in this thesis.The knowledge graph is mapped into vectors by using the representation learning method which introduces the triple context and the entity description text.The personalized recommendation is conducted by utilizing bayesian personalized ranking method.Finally,the two parts are combined by means of joint learning to form an integrated recommendation model.(3)The effectiveness of the proposed model in(2)is verified by experiments,and based on diet knowledge graph a recipe recommendation system is designed and implemented as well.The system employs the recommendation model proposed in(2)to finish personalized recipe recommendation for users on the data constructed in(1).In general,considering the characteristics of the recipe recommendation scene,a personalized recipe recommendation model combined with knowledge graph is proposed in thesis.To meet the requirement of recommending recipes for people,a recipe recommendation system baesd on the model is designed and implemented in this thesis.