Research On Personalized Recommendation Algorithm Based On Food Knowledge Graph

The era of massive information redundancy has arrived,and the problem of data overload has become a problem for human life.The personalized recommendation algorithm can solve this problem effectively by filtering and filtering items that meet users’ individual preferences.However,there is lack of research on recommendation algorithms in the food field,and mainstream food social platforms still use simple interactions such as recipe hotness,adding favorites,and click rate as the basis for coarse-grained food recommendations,which lags far behind the growing personalized needs of people.However,with the continuous development of knowledge graph technology,the powerful computing power of graph neural network makes knowledge graph a powerful tool for personalized recommendation algorithm.By utilizing the rich structural and semantic information in knowledge graphs,user preference information can be better captured,thus improving personalized recommendation precision.This paper makes full use of the advantages of knowledge graph and combines the characteristics of food field to carry out the research on personalized recommendation algorithm for food knowledge graph.The main work is shown as follows.(1)Aiming at the dynamic preference of user demand and the non-atomicity of recipe content in personalized recommendation tasks,this paper proposes a personalized recipe recommendation algorithm based on knowledge perception hierarchical attention network(RMHAT).First,the graph is encoded using the graph attention network,and the weight of each neighbor is spread through the attention mechanism.At the same time,the collaborative information of recipes and ingredients is captured by adding a collaborative graph,and the node features in the food domain knowledge graph are introduced to enrich the semantic relationship,so as to further improve the quality of embedded learning.Fusion node embedding and multi-graph structure embedding to obtain high-quality embedded expression of recipes and ingredients.Finally,the hierarchical attention network is used to distinguish the importance of different ingredients to recipes and the personalized preferences of users for different foods and ingredients.Finally,a multi-layer perceptron is used at the prediction level to predict recipes.(2)Aiming at the problem of data sparsity and coarse granularity of node representation in knowledge graph recommendation algorithm,this paper proposes a personalized recipe recommendation based on cross-view contrast learning(MKCLN).For the personalized recommendation task of recipes,this paper emphasizes the exploration of multi-level perspectives,and makes full use of the higher-order structure information in the heterogeneous graph,as well as the local collaboration information and semantic information,to learn the higher-quality node embedding expression.A self-supervised cross-view comparative learning method is introduced to make the information of multiple views cooperate and supervise each other,and the learning nodes are finally embedded with high quality.Finally,through the joint training and model prediction function,the recipes that meet the personalized preferences are matched with users to complete the recommendation task.(3)Based on this research,a visual prototype system is designed and implemented.Visualize the constructed knowledge graph,and recommend foods that meet the user’s personalized preferences based on the user’s historical interactive information.