The Research Of Community Detection Algorithm Based On Graph Convolutional Neural Network

Community detection on graph-structured data is an important task in the field of data mining and pattern recognition,and it is also the basis for machine learning's Embedding and Classification tasks.Community detection is intended to find potentially relevant connections from complex graph-structured networks,such as the relationships between individuals in social networks,and potential pattern information in graph-structured data.However,due to the increasing complexity of data forms in recent years,traditional community detection methods have shown many limitations including performance and stability.Improving existing community detection algorithms and discovering new forms of community detection algorithms are still one of the problems urgent to be solved.Graph Convolution Neural Network(GCN)is an extension of the traditional Convolution Neural Network(CNN)on high-dimensional graph data.It makes up for the deficiency that traditional neural networks cannot work on non-Euclidean distance data,and makes it possible to use neural network classifiers for efficient parameter sharing on graph-structured data.In this paper,the convolution kernel of graph convolutional network is analyzed in detail and based on its similarity with the information propagation model,it is unsupervised and improved to meet the community detection algorithm requirements.Since the information dissemination model is abstract from the real world,the community detection based on this model should be more in line with the real world results.This paper takes the graph convolutional neural network of machine learning in graph structure as the method and the information propagation method as the reference model,apply the semi-supervised machine learning method to un-supervise domain.Then proposes an unsupervised community detection algorithm based on the graph convolution network.The algorithm modifies the graph convolutional neural network parameter sharing model and uses fixed weights instead of weight training to realize the application of supervised learning methods in the field of unsupervised.And obtain the characteristic information of each node by simulating the propagation of the initial signal layer by layer,and finally obtain the community division of the node by comparison.The algorithm proposed in this paper only uses the topology of the graph structure,does not rely on the original label of the data,and reduces the complexity of the algorithm while obtaining better results.The research of graph convolutional neural network in the application and community detection tasks is studied.The main work completed in this paper is summarized as follows:First,the concept and development of community detection and machine learning are introduced.It also illustrates how machine learning methods can be used in community detection,followed by graph convolutional neural network models that can be used in non-Euclidean distances.And design community detection algorithm based on graph convolutional neural network.Second,Explain in detail the deep learning on the graph,compare the differences between the deep learning methods on the graph and traditional machine learning methods,and further explain the application and problems encountered by machine learning methods in community detection,and machine learning The method may be extended in other areas.Then,The information propagation model on the graph structure is introduced in detail,and the process and working principle of the community detection algorithm based on graph convolution neural network proposed in this paper are compared.Demonstrate that the two models can achieve the same time unit,so that nodes with no information label on the graph data can obtain information.At the implementation stage of the algorithm,it solves practical problems such as the initial selection of analog signals in the algorithm and the possibility that the division results may be abnormal.The algorithm analyzes and solves practical problems such as the initial selection of analog signals and possible abnormalities in the segmentation results.The semi-supervised machine learning method is integrated with unsupervised community detection.The algorithm is applied to a variety of artificial and real data sets to test and verify the test results.The results show that compared with other community detection algorithms,the proposed algorithm has certain advantages in the accuracy and complexity of community detection.Compared with other community detection algorithms with fixed parameters,it has the characteristics of adjustable parameters.Finally,summarized the advantages of the algorithm and the areas that need to be improve.Explains the later expectations and explores several other directions worthy of study.