A Contrast-Enhanced Graph Neural Network Recommendation Algorithm

Abstract

The recommendation algorithm based on graph neural network focuses on encoding the embedded representation of users and items through interactive data, ignoring the interference of uninteracted data, resulting in inaccurate recommendations. To solve the above problems, firstly, the graph convolution encoder is used to generate the vector representations of the users and the items. Secondly, contrastive learning is carried out in each training batch, so that the user vector is close to the interactive item and far away from the non-interactive item in the representation space, and the distribution of the user vector tends to be scattered to alleviate the mutual interference between users. In order to verify the effectiveness of the algorithm, experiments were carried out on the datasets Yelp2018 and Amazon-Book, and the recall rate was increased by 6.07% and 3.35% compared with the advanced model, respectively.