Research On Recommendation Model Of Graph Convolutional Network Based On Subgraph

Graph convolutional networks can learn user and item embeddings using collaborative signals from high-order neighbors,so it is widely used in recommendation systems.But like other graph convolutional models,recommendation models based on graph convolutional networks will encounter over-smoothing problems,that is,when more layers are stacked,the node embeddings will become more similar and eventually become unrecognizable,resulting in recommendation performance decline.In this context,this paper studies the graph convolution network recommendation model,and proposes a graph convolution network recommendation model based on subgraphs.This model solves the problem of excessive smoothing in the graph convolution network recommendation model by high-order graph convolution of subgraphs.The model first establishes interaction graphs for all users and items,then uses random matching algorithm to coarsen the adjacent points to aggregate,convert large-scale graphs into smallscale graphs,and then use recursive multi-level dichotomy to initially divide small-scale graphs,and finally it is refined and restored to the original graph step by step,and the divided relationship subgraph is obtained.The divided subgraph and the original image are used as the input of the graph convolution network,and the high-order and low-order graph convolution operations are performed on the multi-convolutional layer.Therefore,the model can avoid spreading the negative information of the high-order neighborhood to Embedded in learning.Then the original input of the node and the embedding generated by the multi-convolutional layer are weighted and summed to generate the final embedding,and finally the final embedding of the user and the item is subjected to the inner product operation to generate the predicted score.Finally,the subgraph-based graph convolutional network recommendation model is designed and the code part is implemented.On the basis of perfecting the test environment,the performance comparison and analysis with the current mainstream graph convolutional network recommendation model are carried out in the data set Gowalla,Yelp2018,AmazonBook.The experimental results on those data set show that this model is significantly better than the existing recommendation model based on graph convolutional networks.