Dish Recommendation System Based On Knowledge Graph

With the development of China's economy and the improvement of people's living standards,people's eating habits and lifestyle have changed a lot.In recent years,the incidence of cancer in China has shown an obvious upward trend,and one of the important reasons for the high incidence of cancer in China is unreasonable diet.In China's long history of food culture,Chinese food has an irreplaceable position,deeply affecting every Chinese diet.This article investigates the existing software related to healthy diet,such as "Mint APP","Gourmet Master",etc.,and finds that it only gives nutritional value information and nutritional value of calories and other related foods,but cannot be combined with the user's physical health.status,nor describe the dietary status in detail.Therefore,it is difficult for these tools to make accurate and effective recommendations for healthy diet.In order to meet users' reasonable diet and healthy diet needs,this article constructs a Chinese food knowledge graph to describe diet information more comprehensively,and at the same time combines collaborative filtering algorithms and knowledge representation learning to provide users with recommendations that are more in line with healthy diet requirements.The following is the main contents of this paper:Firstly,the knowledge graph of dishes was constructed.This paper analyzes the knowledge characteristics of the acquired recipes and nutrition related literature date,realizes the division of entities and relationships in the field,and defines the relationships among multiple entities;Using Bi LSTM-CRF to extract entities from semi-structured and unstructured text data,and aligns the entities;Finally,the graph database Neo4 j is used to store the constructed knowledge graph of the dish domain.Secondly,a recommendation algorithm combining collaborative filtering and knowledge representation is proposed,which not only uses the user behavior data,but also contains the similarity information of the dishes themselves.By using the knowledge representation learning algorithm Trans D,The entities and rich semantic relationships in the knowledge graph of dishes are accurately mapped into the low-dimensional vector space to generate the similarity expression of dishes based on the semantics;Building user interest model based on information collected from users,obtain user behavior matrix,and generate user-based scoring,the similarity expression of dishes based on user scoring is generated,and the final similarity expression between dishes is obtained by linear fusion of two similarity information,and then combine the user scoring matrix and the fused similarity expression to combine the user scoring matrix to predict the unscored dishes of the users.According to the prediction score in descending order and select Top-N dishes from it as a recommendation list and push it to the user.Finally,by combining the knowledge graph of dishes constructed in this paper and the user's scoring data on the dish,we use the recommendation algorithm proposed in this paper to carry out experiments.The results show that,compared with the traditional collaborative filtering algorithm,the proposed algorithm combined with the semantic information of dishes achieves an accuracy of 79.3% and an F1 value of 76.3%.