Recommendation systems have become an integral part of various online platforms, providing personalized recommendations to users while reducing information overload. Graph Convolution Neural Network (GCN) based models have become the most popular research method in recommendation systems. These models have the ability to learn embedding representations on graph structures, which makes them effective in modeling recommendation systems. However, the GCN model suffers from the oversmoothing problem, which makes node representations indistinguishable after multiple layers of graph convolution. Researchers have proposed models such as LR-GCN, Light-GCN, DropEdge, and IMP-GCN to mitigate this problem. Despite the significant advancements in GCN-based models, they still do not take full advantage of higher-order neighbors' covariance signals. Additionally, they do not consider the influence of users' different opinions on the recommendation results when propagating and aggregating synergistic signals. This paper proposes a graph convolutional neural network recommendation model with interest-aware message delivery that utilizes 2nd-order neighbors' cooperative signals and users' views to improve accuracy in learning node embeddings. Our model separates subgraphs for users and products, allowing for intergroup crossover in the classification. The proposed model addresses the shortcomings of current GCN-based models, and it is evaluated on several datasets. The results show that our model outperforms state-of-the-art models in terms of prediction accuracy, making it a promising recommendation model for various applications.