Research On Community Detection Technique Based On The Fusion Of Nodes And Graph Structure

Network structure is ubiquitous in daily life,from the social network and the Internet to the food web and virus transmission network,etc.Developing a comprehensive understanding of network structure is helpful for maximizing the utilization of networks and deal with the risks and challenges brought about by them.Graph clustering is a typical problem in network structures,the purpose of which is to group together nodes with equivalent natures or high similarities.An instance of this is community detection,which is essentially a method for finding closely-linked nodes within a network.The continuous technical development of graph neural network over the recent years has provided a new direction for graph clustering.Graph neural network facilitates the mapping of nodes into eigenvectors,and comprehensively considers the characteristics of the network’s structure and nodal features.The obtained eigenvectors can describe the characteristics of nodes well.In light of these benefits,this paper employed graph neural networks to perform community detection.The main research contents and innovations are as follows:(1)In order to solve the problem that traditional models need to specify the number of communities in advance when dividing communities,we proposed a model based on GraphSAGE and designed an algorithm to detect communities in the case of unknown number of communities.The model is composed of GraphSAGE layer and full connection layer to judge whether node pairs belong to the same community.The algorithm uses the model we proposed to judge whether nodes can join a certain community and divide them by iteration.And we designed experiments to verify the effectiveness of the model and the algorithm.Experiments show that our model can accurately determine whether nodes belong to the same community,and the algorithm can accurately divide the community structure in the network.(2)In order to explore the ability of our proposed algorithm to divide communities in dynamic networks,we designed experiments to verify.During the experiment,snapshot-based method was used.We use The network structure of the previous moment to train the model,and then the trained model is used to divide the complete network.Experimental results show that our algorithm can accomplish the task of community division in dynamic network.(3)This paper also improved the CommDGI model and designed a new encoder that comprises a two-layer graph convolutional neural network.The multi-layer feature fusion method was then used to enhance the encoding capability of the encoder.To verify the effectiveness of the model,experiments were conducted using three real data sets:Cora,CiteSeer and PubMed.The experimental results show that the improved model is more accurate than the comparison model.