Research On Important Nodes Identification And Community Detection In Mobile Social Networks

As the cross network of mobile communication networks and social networks,mobile social networks(MSNs)support data or information transmission between mobile users.On the one hand,the general existence of power law distribution makes the importance of users different;On the other hand,the ubiquity of community structure makes the similarity of users’ manifest.However,how to effectively use the data information to identify the important nodes and detect communities in such a particular scenario are still not been effectively solved.Therefore,the research of this thesis focuses on the important nodes identification and community detection in mobile social networks.Ulike the rest nodes in the network,important nodes may often affect the network structure,and even exert a greater influence on the network function.In addition,they also play a decisive role in the process of information propagation.The importance sequence of nodes is the sequence of nodes that is generated by ranking them by the relevant metrics.Important nodes tend to attract other nodes to attach to them and thus form communities.A community is a structure that lies between a macro network and a micro node,characterized by tight internal connections and sparse external connections.So understanding its structure contributes to its detection.1.In view of the problem that existing important nodes identification methods in mobile social networks do not fully consider the activity of nodes,which leads to the failure of accurate identification of important nodes,an important nodes identification method based on activity degree is proposed.This method defines the activity degree of nodes from the perspectives of topology structure and behavior characteristics,and the nodes with higher activity degree are chosen as seed nodes.Then we simulates the process of information propagation by using the SIS(suspected-infected-suspected)model,and the propagation ability of seed nodes is measured.Through the analysis of the experimental results on the four real datasets,it can be found that compared with the existing centrality based method and optimization based method,the proposed method can ensure that important nodes can be identified more effectively without increasing complexity.2.In view of the problem that the existing evaluation methods of node influence in mobile social networks do not fully consider the topological structure and social attributes,which leads to the relatively one-sided evaluation of social influence,an important node identification scheme based on topological structure and social attributes is proposed.This method gives the definition of similarity between nodes and the improved constraint value of node network.On this basis,we define the social influence of nodes,and the nodes with higher social influence are chosen as seed nodes.Then,this thesis simulates the process of information propagation with the help of the SIR(suspected-infected-recovered)model,and the propagation ability of seed nodes is measured.Through the analysis of the experimental results on the four real datasets,it can be found that the proposed method can identify important nodes more accurately than existing methods.In addition,the effect of nodes on the network robustness is investigated.3.Aiming at the problem that the existing community detection algorithms in mobile social networks ignore the influence of the social attributes and behavioral characteristics of nodes on the formation of communities,which leads to the low accuracy of the division of some communities,a community detection algorithm based on local density clustering is proposed.The algorithm maps the nodes in the network to the data points in the clustering algorithm from three perspectives,namely,the topology structure,social attributes and behavioral characteristics of the nodes.Accordingly,the values of the nodes in the above three perspectives are mapped to node features in the corresponding three dimensions.On this basis of that,community detection is performed with the help of local density clustering algorithm.Through the analysis of experimental results on four real datasets,it can be found that the proposed method can effectively detect communities,and it also has certain advantages in two metrics of modularity and normalized mutual information.