Research on mesh node classification algorithm based on improved heterogeneous graph neural network

In the current era of big data, graph data structures have been widely used in social networks, knowledge graphs and recommendation systems because of their ability to effectively represent complex relationships. However, processing Heterogeneous Graphs with multiple types of nodes and edges remains a challenge. The existing Heterogeneous Graph Neural Networks (HGNNs) have solved this problem to a certain extent, but there is still room for improvement in feature aggregation and cross-type information fusion. In this paper, we propose an innovative neural network model for heterogeneous graphs. First, we introduce a Multi-Channel meta-path Aggregator, which captures complex relationships between different types of nodes in heterogeneous graphs by defining different Meta paths. Each meta path corresponds to an independent information propagation channel, so that the model can learn node features from multiple perspectives and enhance the understanding of the graph structure. Second, to better handle different types of nodes and edges, we designed an Adaptive Attention Mechanism. This mechanism allows the model to automatically adjust its attention weight based on the specific type of node and edge, thus capturing important information more accurately and reducing the impact of noise. Further, we propose a Hierarchical Feature Fusion Module to solve the fusion problem between features of different types of nodes. This module combines low-level local features with high-level global features to form a richer node representation. In this way, the model is not only able to identify subtle differences in local structures, but also to grasp important patterns across the entire layer surface. The experimental results show that our improved model still shows significant advantages in node classification tasks on multiple public heterogeneous graph datasets.