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

Traditional recommendation algorithms are often plagued by data sparseness and cold start problems.The introduction of knowledge graphs and the development of graph neural network(GNN)technology have brought new ideas for solving these problems.Because the knowledge graph contains a lot of human knowledge information,and the potential correlation between the information and it also has a good scalability,and the GNN method can extract these potential information well.The existing recommendation algorithm based on GNN and knowledge graph uses GNN to process the knowledge graph and introduce it as auxiliary information into the recommendation system,which can effectively solve the problems of cold start and data sparseness.However,these methods are relatively simple in the neighborhood sampling and neighborhood aggregation design,and ignore the user information that is also important in the recommendation system,resulting in a general recommendation effect.This paper firstly proposes a sampling strategy based on importance and an aggregation strategy based on pooling operation for the neighborhood sampling and aggregation.This sampling strategy measures the importance of different neighboring nodes through the calculation of the closeness of the relationship,and can more targetedly select the neighbors that have a greater impact on the target project node.The aggregation strategy,by introducing a trainable pooling layer and training neighborhood aggregation weights to aggregate neighborhood features,can more differentially aggregate the impact of different neighbors on the current node.On this basis,an improved algorithm KGCN-PL combining these two improved strategies is proposed.In addition,the existing GNN-based knowledge graph recommendation model usually only considers the one-sided information on the project side,ignoring a large amount of related information between users.To address this issue,this paper proposes a user-side information-based GNN recommendation model UVGNN-U,which recommends users by capturing their potential interests and fusing them with the KGCN-PL algorithm proposed in this paper A GNN knowledge graph recommendation model UVGNN based on two-sided information to further improve recommendation performance.Finally,the proposed algorithm model and the baseline model are compared and evaluated in detail on 6 real-world data sets to verify the effectiveness of the proposed improvement strategy and model.