A Study Of Analysis And Optimization Of Complex Network Structure Based On Swarm Intelligence Algorithm

The Internet has a huge impact on real life,and the value generated by the data is far beyond imagination.“Internet of Everything” links more and more data together,making the network more relevant and more valuable.Complex network model is a useful way to model connection data,and the qualitative and quantitative research of complex network has promoted the development of system dynamics,social psychology,biological sciences,computer science and so on.Therefore,complex network has become an important tool to analyze network connection data,and it has been widely concerned by researchers.The studies of complex network,such as the community structure,the network robustness,the network propagation,influence maximization and structural balance,are useful to understand,predict and control the potential system functionality.When modeling the connection data of real life with complex network model,the problems can often be solved by the optimization method.In particular,the optimization problem modeled by network model is similar to the combinatorial optimization problem,and both of them are NP-hard problems.The thesis mainly focuses on the control of network spreading,influence maximization and structural balance analysis in complex networks.The traditional optimization method can not get the optimal solution when solving NP-hard problems,and the method based on greedy searching is inefficient and difficult to find the global optimal solution.However,the swarm intelligent algorithm has advantages in solving NP-hard problems.The main works of this thesis are summarized as follows.Epidemic threshold of the network,which fundamentally depends on the network structure itself,is a critical measure to judge whether the epidemic dies out or results in an epidemic breakout.Epidemic threshold is regarded as the objective function to control the spreading process.In addition,an efficient structure optimization strategy based on memetic algorithm is proposed to adjust the spreading threshold without changing the degree of each node.Lowering the threshold can promote the spreading process whereas heightening the threshold can prevent the spreading process.In the proposed algorithm,genetic algorithm is adopted as the global search strategy and a modified simulated annealing algorithm combined with the network is introduced as the local search method.Results on artificial and real-world networks demonstrate that our algorithm has better performances for both the threshold minimization and maximization problems.Influence maximization in social networks aims to find a small group of individuals,which have maximal influence scope.We introduce a local influence index as the objective function.The local influence index can provide a reliable estimation for the influence propagations in independent and weighted cascade models.A discrete particle swarm optimization algorithm is then proposed to optimize the local influence criterion.The representations and update rules for the particles are redefined in the proposed algorithm.Moreover,a degree based initialization strategy and a local search strategy are proposed to accelerate the convergence.Experimental results on four real-world social networks demonstrate the effectiveness and efficiency of the proposed algorithm for influence maximization.A novel bi-objective model is presented for social network structural balance,and a multiobjective discrete particle swarm optimizer is used to optimize the bi-objective model.Each single run of the proposed algorithm will yield a set of Pareto solutions,and each solution represents a certain network partition.Consequently,by simultaneously optimizing the objectives in the proposed model,one may have many choices to analyze the balance problem.Extensive experiments compared against several other models and algorithms have been done.All the experiments indicate that the proposed model is helpful for social network structural balance analytics,and that the algorithm is effective.