Graph Representation And Label Propagation Based Community Detection And Its Applications In Complex Networks

The research of complex networks has penetrated into physics,engineering,computer science,biology,economics,sociology and other fields.The detection and analysis of community structure in complex network is helpful to further mining and analyzing the interaction between individuals in complex system,the potential role of individuals in the whole system and attribute characteristics.Based on the thinking of how to better reflect the actual structure of the network,how to further deal with large-scale network and overlapping community detection and how to better use the community structure information in complex network,graph representation and label propagation based community detection and its applications in complex networks are studied.This dissertation has carried out relevant research and some research results as follows:(1)A complex network graph embedding method based on the shortest path matrix and MOEA/D for community detection is proposed,which can better reflect the network structure at the level of network community structure.Firstly,by calculating the shortest path matrix between nodes in the network,the node relationship matrix is obtained by adding the node similarity.Next,aiming at the problem of community detection in disconnected networks,a decomposition-based multi-objective optimization method is proposed to assign distances to unrelated nodes.Then,the network similarity matrix is calculated based on the relationship matrix of network nodes,and the low-dimensional vector representation of nodes is obtained by random surfing strategy and multi-dimensional scaling method.Finally,the community structure of the network can be detected based on the obtained node representation structure.Starting from the essence of network structure and the tightness between nodes,this method can reflect the relationship characteristics of network nodes more effectively,and then obtain the vector representation of nodes which can more accurately reflect the information of community structure in networks.The test results on 11 networks show that the node vector representation results obtained by this method can better reflect the community structure information in complex networks.(2)A large-scale community detection algorithm based on core node and label propagation layer by layer is proposed.Firstly,the core nodes whose node degree is greater than the average node degree in the graph are found.This can effectively utilize the feature that the core node is a potential community center,and avoid the impact of nodes with low node degree on community structure detection.Then,starting from the core node,label propagation is carried out layer by layer according to the node degree and node connectivity,which can effectively improve the accuracy of community detection.The algorithm also calibrates the node labels after the label propagation according to the attraction degree of the current community to the nodes,which effectively improves the misclassification in the early community detection process.Finally,overlapping communities are detected on the basis of non overlapping community structure to make the detection results more accurate and interpretable.The results of community structure detection on 13 real datasets show that the proposed algorithm has more advantages than 4non overlapping community detection methods and 2 overlapping community detection methods.(3)A Graph convolutional network training algorithm based on community detection method is proposed to obtain higher accuracy.Firstly,the graph is divided into several sub-graphs by using community integration strategy.Then,GCN samples are limited within these sub-graphs in each step to reduce redundant neighborhood search.In order to make the structure information of sub-graphs more suitable for GCN training,an improved modularity increment function is proposed as the objective function of community integration strategy by emphasizing the relationship between communities.In order to prevent GCN sampling from losing information due to the limitation of objective function,the second update strategy of label is added in the process of community integration.This makes the direction of community integration more diverse,and effectively reduces the generation of local optimal situation.In addition,it is not necessary to set the number of sub-graphs in advance for networks of different sizes and structures.Finally,the GCN training results are tested by the node classification task.Compared with several state-of-the-art algorithms,the proposed algorithm can get higher accuracy in community detection and node classification.(4)A rating prediction algorithm based on community detection and graph neural network is proposed.Firstly,in the process of community detection of user-user graph,a pre-partition based on label propagation algorithm is added to ensure that users with closer ratings of the same item are more likely to be divided into the same community.This not only satisfies the principle that users with similar preferences are similar,but also increases the interpretability of the rating prediction in recommendation system.Then,community integration strategy is used to detect the community structure of user-user graph,and similar user subsets are obtained.These subsets can reflect the similarity between users and make full use of the information in user-user graph.Finally,the final rating prediction result is obtained by using graph neural network.A new loss function is proposed to train the model,which can alleviate the problem that the common loss function is easily affected by outliers and reduce the overall performance of the model.Compared with the 5state-of-the-art algorithms,the proposed algorithm can improve the rating prediction ability of the model on the three social network datasets.(5)A dynamic node immune model,based on community structure and threshold is proposed.Firstly,a network model is established,which regards nodes carrying harmful information as new nodes in the network.The method of establishing the edge between the new node and the original node can be changed according to the needs of different networks.The propagation probability between nodes is determined by using community structure information and a similarity function between nodes.Secondly,an improved immune gain,based on the propagation probability of community structure and node similarity,is proposed.The improved immune gain value is calculated for neighbors of the infected node at each time step,and the node is immunized according to the hand-coded parameter: immune threshold.This can effectively prevent invalid or insufficient immunization at each time step.Finally,an evaluation index,considering both the number of immune nodes and the number of infected nodes at each time step,is proposed.The immune effect of nodes can be evaluated more effectively.The results of network immunization experiments,on eight real networks,suggest that the proposed method can deliver better network immunization than several other well-known methods from the literature.To sum up,the community structure information of complex networks is studied in this dissertation.A community detection method based on graph embedding,a large-scale community detection algorithm based on core node and label propagation layer by layer are proposed.Furthermore,the community structure information of complex system,the interaction between individuals and the potential role of individuals in the whole system are mined and analyzed,and the application of community detection method is studied.A graph convolution network training algorithm based on community detection,a rating prediction algorithm based on community detection and graph neural network and a dynamic node immune model based on community structure and threshold are proposed.