Recommendation Algorithm and Simulation Experiment Based on Graph Representation Learning

Abstract

Based on the GAN construction, the recommendation model in this paper preprocesses the interaction score between users and items, mines the implicit relationship embedding vector representation between users and items according to the user-item interaction bipartitic graph, extracts the explicit relationship between users and items from the graph, fuses the implicit relationship vector of nodes with the explicit relationship vector, and calculates the predictive score to obtain the recommendation result. Emphasis is placed on recessive relation extraction, dominant relation extraction and similarity calculation. The proposed model was verified by simulation experiments, including the construction of experimental environment, the selection of data sets, the design of evaluation indicators, the simulation experiment, and the analysis of experimental results. Compared with the best baseline model KGAT, the NDCG@5 index is improved by 2.31%. Compared with GraphRec, the worst baseline model, the RMGAT model improved the NDCG@5 index by 25.49%.