Graph Neural Network Model Based on Layered Converter Aggregation

Graph data is an important data in real society. With the development of neural network technology, graph neural networks are receiving more and more attention. However, the existing methods mostly use adjacency matrix or attention to aggregate the information of surrounding nodes, or use random walk to select neighboring nodes for aggregation, but this aggregation method is too simple and ignores the complexity of the graph structure. At the same time, inputting all the massive graph data into GCN for large-scale matrix operations in the real world requires a lot of computational power, and the attention calculation between multiple nodes also brings a heavy burden to computers, resulting in additional time loss. Based on this, this article proposes a hierarchical transformation aggregation model to aggregate node information by hierarchy and region. To put it simply, we first select nodes in each layer in the way of random walk, and then use GCN to aggregate. After obtaining the node characteristics of each layer, we use the attention mechanism with transformation to aggregate the node characteristics of each layer to obtain the final representation of nodes. Finally, we verify the reliability of the model on multiple datasets.