Research On Recommendation Algorithm Based On Knowledge Graph And Graph Neural Network

The wide application of Internet technology has spawned the field of data mining,including the rapid development of data mining fields such as knowledge engineering and big data decision-making.On the one hand,as a highly developed product of the Internet,knowledge graph is also an excellent knowledge base that has attracted more and more researchers’ attention.It stores information in the objective world in a form of structured representation that conforms to cognition,mapping an entity or abstract concept in the real world with its attributes into nodes and edges between nodes on the graph structure.The essence of the knowledge graph is a multi-source heterogeneous network,which can store a large amount of cross-domain information.On the other hand,traditional recommendation systems inevitably face the problem of cold start in the face of more real-time and cross-domain scenarios.In this context,the recommendation system based on knowledge graph assistance has more advantages than the traditional recommendation system,mainly reflected in: accurate,efficient and explainable.Most of the recommended methods of existing studies use knowledge graph representation learning for information mining of entity features,but some of them ignore the neighborhood information of knowledge graph entities,resulting in the loss of highorder feature information.In addition,some methods do not fully consider the heterogeneity of the knowledge graph,and the representation of the relationship of the knowledge graph is uniformly modeled with the relationship vector.Considering the above problems,the main contents of this paper are as follows:(1)Aiming at the problem that the existing recommendation method ignores the overall information of the knowledge graph entity information mining process,a recommendation method based on graph neural network enhanced neighborhood information is proposed.The graph attention network is introduced to carry out neighborhood information aggregation to make up for the neighborhood information of missing entities in the process of extracting the features of the knowledge graph in existing research.In this study,the comparative experiments,statistics and analysis of click-rate prediction and top-K recommendation prediction were carried out on the data under the four backgrounds of movies,books,online courses and music,and the experimental results showed that the method improved in a number of indicators,which confirmed the effectiveness of the method.(2)Aiming at the heterogeneity of knowledge graph,a recommendation method based on hierarchical learning of graph neural network is proposed,which realizes a more finegrained modeling of information propagation between nodes.The knowledge graph is layered by the relationship type in the triplet data,and then the graph neural network is used to simulate the information propagation behavior on different semantic relationships.Secondly,in the recommendation task,the attention mechanism is introduced to calculate the attention distribution of recommended items in the user interaction history,simulate collaborative filtering,and mine user preferences.Finally,under the background of film and online course data,this study verifies the recommended performance of the method,and experiments,analyzes and discusses the influence of some key parameters and structures on the model.