Graph Neural Network Classification Algorithm Based On Graph Sampling

In recent years,Graph Neural Networks(GNNs)have been successfully widely applied in some important fields such as biomedicine,system modeling,recommendation systems,text classification,and social networks.With the rapid growth of artificial intelligence,the graph-structured databases have become increasingly large scale.Training on the large-scale graph-structured databases imposes challenges for graph neural networks.In this paper,a random walk graph sampling algorithm is first used to generate the multiple batches of subgraphs in a random walk way,which can reduce the complexity in training on the large-scale graph database to enhance the generalization of the deep model.On this basis,the Random Walk Graph Sampling based Graph Convolutional Networks Classification(RWGS-GCNC)methods is proposed to integrate the features of high-degree nodes in the graph to extract the relevant semantic information between the long-distance nodes.Furthermore,RWGS-GCNC enhances the perception of whole graph-structured data by the multiple batches of subgraphs mechanism.Finally,experiments on five open graph-structured databases demonstrate the effectiveness of our proposed methods in the inductive learning scenario.In order to obtain the various effects of nodes on extracted features,the Random Walk Graph Sampling based Graph Attention Networks Classification(RWGS-GATC)is proposed to learn the attention coefficients and the gate weights of the multi-head by the multi-head attention mechanism.Hence,the feature of the graph-structured data can be dynamically extracted to yield improvement in the classification accuracy.Finally,multiple experiments were conducted on five open databases.The results show the proposed method outperforms the state-of-the-art methods,verifying the advantage of RWGS-GATC in inductive learning.