Recipe Data Knowledge Graph Recommendation System

With the vigorous development of the Internet and various information technologies,the world has transitioned from an era of lack of information to an era of excessive infor-mation,prompting more and more researchers to focus on the research of recommendation systems,hoping to make use of its powerful functions.The recommendation system helps people automatically filter information,thereby reducing the cost of information screen-ing.In this thesis,we will focus our research on specific dietary areas.Food,clothing,housing and transportation are the four main areas of people's daily consumption.Among them,”diet” ranked first,enough to explain its importance.In order to meet users' dietary decision-making needs and guide users to cook,various recipe websites have emerged.They collect recipes carefully and spend a lot of time verifying the reliability of the recipes.However,the biggest problem that users encounter when browsing recipe websites is that it is difficult to quickly obtain content that suits you in the face of a large number of recipes.As a site with a large amount of recipe content and user information,recipe websites sel-dom fully excavate the recipe itself and the historical behavior of users,and thus cannot provide users with personalized recipe suggestions.In response to this problem,this thesis proposes the RKGR model(Recipes Knowl-edge Graph Recommendation Model).Based on the establishment of a recipe knowledge graph,this model is used to mine the potential associations between recipes and user pref-erences to meet the needs of personalized recommendations on recipe websites.The main tasks of this thesis are:(1)Construct the knowledge graph in the field of diet.By crawling a large amount of recipe data and user history information from foreign recipe websites,and through the steps of ontology construction of knowledge graph,deep learning knowledge extraction,Bi LSTM-CRF algorithm for recipe feature supplement,a recipe knowledge graph with rich entities and relationships is constructed,which provides data basis for personalized recommendation in the diet field?(2)A recommendation algorithm based on recipe knowledge graph was constructed.The algorithm fully mines the user's historical evaluation data and published recipes,ex-tracts the entities and constructs subgraphs,uses the multi-channel convolution technology based on natural language to model the user's preference,and finally uses the deep neural network to get the probability that a candidate recipe will be liked by the user?(3)The superiority of RKGR model is verified by experiments.On the data set of recipe knowledge graph,we designed four groups of different comparative experi-ments,and analyzed the experimental results to verify the scientificity and rationality of the model.To sum up,this thesis provides a new idea to solve the problem of low accuracy of recipe website recommendation effect,which helps the website to recommend recipes in line with its own preferences to users in a faster and more accurate way,so as to meet the personalized needs of users.