Research Of Personalized Recommendation Technology Based On Knowledge Graph

In the era of the rapid development of Internet and the explosive growth of information,people are prone to information overload and other problems,and it is difficult to obtain effective information and learn knowledge.In order to solve the problem of information overload,personalized recommendation system came into being.Compared with search engines,personalized recommendation systems are more suitable for application scenarios where user needs are unclear or cannot be accurately described by keywords,such as our common news information,e-commerce,and music applications.In practical applications,most recommendation systems generally use collaborative filtering recommendation algorithms,only input user interaction data,there are sparseness problems and cold start problems,which to some extent limit the recommendation effect.From the root,the imperfection of data is the root cause of data scarcity and cold start.Therefore,this paper introduces knowledge graph as auxiliary information,focusing on the distributed representation method of knowledge graph and personalized recommendation algorithm based on knowledge graph.The main research contents of this article are as follows:Firstly,for the distributed representation method of knowledge graph,this paper introduces the problem that the existing graph distributed representation method loses the high-order similarity of subgraph level.To this end,this paper proposes a knowledge graph distributed representation model KG-GRU based on recurrent neural network,using sequences containing nodes and relationships to model the similarity of subgraphs,and representing relationships and entities in the same embedding vector space.In addition,this paper proposes a jump or stay strategy to guide random walks to sample the data of knowledge map,which avoids the problem of manual construction of meta path and unbalanced distribution of node types.Secondly,this paper proposes two personalized recommendation algorithms based on knowledge graph: KG-CF and KG-GRU4 Rec.Based on the idea of collaborative filtering algorithm of fused content,KG-CF directly merges the distributed representation vectors of items in the domain knowledge graph into the calculation of item similarity,and adds the semantic information of items to the traditional item-based collaborative filtering algorithm,thereby improving Personalized recommendation effect.KG-GRU4 Rec improves the knowledge graph distributed representation model KG-GRU proposed in this paper,and implements an end-to-end model for predicting user ratings,avoiding the problem that KGCF's score prediction still depends on user historical rating data.Finally,in the experimental stage,the movie recommendation is chosen as the application scenario,which is widely used in the field of personalized recommendation.In order to evaluate the above-mentioned algorithm model proposed in this paper,this paper investigates and implements the construction of the knowledge graph in the movie domain,including from the construction of the ontology database in the movie domain,the crawling of movie-related data,to the extraction and storage of knowledge.Finally,this article proves that the KG-GRU model can learn the more accurate distributed representation vector of entities and relationships in the movie knowledge graph through the link prediction experiment.Through the top-N movie recommendation experiment,it proves that KG-CF and KG-GRU4 Rec recommendation algorithm are better than the comparison algorithm in hit rate and average reciprocal ranking.