Light Graph Convolution Neural Network(LightGCN)is a hot research topic in recommendation systems.It applies deep learning to recommendation systems and simplifies the graph convolution neural network applied in recommenda-tion systems,greatly improving the performance of recommendation.However,the loss function in LightGCN is relatively simple.Only BPR loss and L2 regularization loss are used,and the dataset is not fully utilized.In this research,it is proposed to integrate contrastive learning into the LightGCN model.Specifically,a boundary is set to filter information with low similar-ity,and super parameters are used to control the relative weight between positive and negative samples,maximize the similari-ty between positive sample pairs,and minimize the similarity between filtered negative sample pairs.This model makes the ex-periment better.The experimental results show that the evaluation index has been improved compared with LightGCN when the experiment is conducted on the same data set under the same experiment settings.
维普
万方数据