Non-Negative Sampling Attention-Gated Graph Convolutional Network for Recommendation

To address these issues that linear propagation cannot accurately handle users and items with diverse feature representations, alongside the instability of the negative sampling strategy, this study proposes a Non-negative Sampling AttentionGated Graph Convolutional Network (NSAG-GCN) for recommendation. This method utilizes a non-negative sampling optimization approach, enabling fast learning of model parameters from all data samples. Moreover, in the propagation process of the GCN network, a gate module based on attention networks is employed to choose different propagation methods. This design not only effectively reduces the impact of data sparsity on recommendation performance but also maximizes the exploration of node correlations and information, enhancing the reliability and stability of recommendation performance. Extensive experiments were conducted on four public datasets, and the results demonstrate that the NSAG-GCN framework outperformed the baseline methods.