一种基于图卷积网络与 Inception 结构的半监督学习模型

In many real-world scenarios such as a citation network or a knowledge base formed by a collection of documents, the data has a graph structure. For such data with graph structure, there is often a problem of node classification on the graph, such as the classification of each document in the document set or the classification of each named entity in the knowledge base. For data with a graph-like structure in these problems, graph convolutional network is a popular neural network method in recent years. Graph convolution network is a generalization of convolutional neural networks on graph structured data and has hierarchical characteristic similar to graph convolution network. This paper uses the ideas in convolutional neural networks to study the network structure optimization of graph convolutional networks, and proposes a semi-supervised learning model based on graph convolutional networks and Inception structures. Inception-GCN (Inception Graph Convolution Network) model. The Inception-GCN model based on the GCN model proposed by Kipf et al and uses the idea of the Inception network structure from CNN to improve the GCN. It has achieved good results in the semi-supervised classification task. It improves accuracy rate by 0.7% compared with the original model On the NELL data set which is used at the proposal of the original graph convolution network model.