Enhanced Attention-Based Graph Convolutional Networks: Boosting Collaborative Filtering Recommendations

Recommendation algorithms can filter personalized information from massive data, effectively alleviating the current social problem of information overload. This study introduces a graph convolutional network with enhanced capabilities collaborative filtering recommendation model combined with graph attention ( Enhanced Attention-based GCN, EAGCN), aiming to address the shortcomings of existing graph convolutional network-based recommendation models, such as the inability to effectively distinguish the impor tance of adjacent nodes during node information aggregation. To mitigate the over-smoothing issue, a residual-enhanced graph convolutional network collaborative filtering recommendation method (R -EAGCN) is designed based on the original model. Comparative experiments between EAGCN, R -EAGCN and the best baseline models show that their recall rate and NDCG metrics outperform other baseline algorithms. Specifically, the recall rates of EAGCN and R -EAGCN are increased by nearly 10% and 15% respectively compared to the five baseline models.