Research On Community Detection For Networks And Limited Resource Allocation Strategies

The society can be viewed as consisting of different networks which are connected with each other inside and outside.Typical examples include technological networks such as the Internet,the World Wide Web,transportation networks,and the resident's mobile travel networks,biological networks such as brain networks and protein-protein interaction networks,and social networks such as criminal networks and acquaintance networks.These networks give the characterization of the real world we are living in,from different perspectives.Therefore,the network science provides the powerful tool for analyzing the common feature of complex systems in real world.For example,the study on community structure connecting among nodes can discover the internal organization of nodes and detect the potential function of networks.In addition,the study on limited resource allocation is able to maximally satisfy different kinds of demands and restrict harmful propagation progress.In the field of community detection,one of the important problems is how to discover potential community structures based on different types of networks.Besides,another important problem is how to allocate limited medical resources during the emergency situation.Revolving around the two aspects,this thesis focuses on two points: “How to efficiently discover community structures” and “How to effectively allocate the resources” which are based on the network science:(1)How to efficiently discover community structures: This thesis focuses on static and dynamic networks based on the network structure and improves the coverage speed of ACO and the robustness of PSO as well as captures the nonlinear representation of dynamic networks.Firstly,the community detection problem in static network is changed into the single objective problem and this thesis uses the ACO method to solve the changed problem.It proposes the novel adaptive control population framework by estimating the status of ants in order to control the number of populations in ACO.After that,the community detection problem in dynamic network is changed into the multiplex objective problem and this thesis uses the evolutionary framework to consider both the quality of community structures and the stability.It proposes a novel PSO to improve the diversity of particles by using the label propagation and genetic operations.Finally,this thesis proposes the autoencoder for detecting dynamic community structures by adopting the evolutionary framework.It improves the robustness of community structures by mapping nonlinear representation characters.The three algorithms are verified in terms of accuracy,robustness and solution quality in synthetic and real-world datasets.(2)How to effectively allocate the resources: This thesis proposes two objectives of high-risk and severity priorities in order to provide the reasonable allocation strategy for Shenzhen center in terms of disease control and prevention.Firstly,the decision-making problem is changed into the multi-objective optimization problem and this thesis uses the novel PSO to solve this problem.After that,this thesis balances the heterogeneity of a highrisk population and the severity of infection and eventually generates a reasonable allocation strategy of medical resources.Finally,taking the megacity Shenzhen in China as an example,experiments show that GA-PSO effectively balances different objectives and generates a feasible allocation strategy in every district.Overall,based on the network science,this thesis uses the swarm intelligence and autoencoder theory to detect community structures in static and dynamic networks.In addition,it proposes the reasonable strategy for allocating medical resources by Shenzhen data.Massive experiments show the validity of proposed algorithms.