Research And Application Of Deep Graph Neural Network

Deep learning methods are becoming increasingly powerful in solving various challenging artificial intelligence tasks.Among these deep learning methods,convolutional neural networks have demonstrated promising performance in many grid-likedate-related applications,such as image recognition and computer vision.In many real-world applications,the data can be naturally represented as graphs,such as social,citation,and biological networks.Graph Convolutional Networks(GCN)is a model that uses deep learning to implementgraph-related tasks.It can fuse the node’s content information,connection information,and label the unobserved nodes.However,this model has the disadvantage of involv-ing complex matrix operations and insufficient explanatory power.In view of the above prob-lems,this paper has improved the model based on previous studies.The main research work is as follows:1)In LGCLs,the neighborhood of the node is the the direct neighbor or indirect neighbor of the node.This model can expand the information of nodes,which may add redundant nodes to neighborhood or ignore key nodes.To solve this problem,we propose determined neighborhood model for constructing a neighborhood of a certain size and define a convolution operation on the neighborhood.Our model different from previous models which simply use the direct or indirect neighbors of the nodes as neighbors.Our model add the direct neighbors and indirect neighbors of the node to the candidate neighborhood of the node,and then sort the nodes in the candidate neighborhood according to the degree of nodes.After sorting,add nodes with higher degree to the neighborhood.Our experimental results on graph-related tasks demonstrate that our model can construct an excellent neighborhood which can reserve the valid information2)In MixHop,the connections between nodes in the neighborhood of node are ignored To solve this problem,we propose ordered neighborhood model which create ordered neighborhood of node by giving weights to the nodes in determined neighborhood of node.First,construct the determined neighborhood of node by using determined neighborhood model.Then,calculate the similarity matrix between nodes by using the feature vectors of the nodes in determined neighborhood.Last,construct ordered neighborhood by using similarity matrix.Our experimental results on graph-related tasks demonstrate that our model can maintain the order structure in the process of training.3)We implement a medical data classification system based on depth map network based on the proposed algorithm.We construct the evolutionary graph of patient by analyzing the medical data of patients in different time periods.We improve the recognition accuracy of the system by applying evolutionary information to the process.