Research On Graph Neural Network With Graph Embedding Model For Session-based Recommendations

At present,Markov chain and recurrent neural network are mainly used in the problem of session-based recommendation system.However,the former is suitable for short sequence data and has the independence assumption,while the latter has the problem of cold start.Sequence data can be transformed into graph structure data,while graph neural network can process complex graph structure data,and overcome the problems of independence assumption and cold start,therefore researchers proposed to use graph neural network model to improve the recommendation performance.In this paper,an improved graph neural network recommendation model,graph neural network with graph embedding model,is proposed innovatively.This model constructs graph structure data according to user's session information,inputs graph structure data into graph embedding algorithm and graph neural network model for training,then obtains embedding vector representation of the commodities,calculates recommendation score of each commodity through attention mechanism,and finally obtains recommendation ranking of the commodities.This model mainly includes two stages:graph embedding and graph neural network.The attention mechanism can dynamically adjust the recommendation solution and make targeted recommendation.In addition,the preprocessing of session information can filter out the noise data and further improve the recommendation performance.The graph embedding stage aims to find the embedding vectors of commodity nodes to represent the low-level features of the commodities.The graph neural network can mine hidden relationship in the graph structure and find the transfer characteristics of the commodities.In this paper,Deep Walk is used as the graph embedding algorithm,which is outstanding in the graph structure data and its downstream machine learning tasks.It has great flexibility in the selection of graph neural network model and gated graph neural network is used in this paper.This model combines the two to better evaluate the transfer relationship of commodities and predict the walk of commodity click sequence in the graph structure.In this paper,experiments are conducted on two datasets,Yoochoose and Diginetica.The experiment results show that this model has better performance in the field of session-based recommendation system,and the recommendation performance is significantly improved compared with the current model in the same field.