Structure Analytics Of Big Data Complex Networks Based On Swarm Intelligence

Represented by iPhone, iPad, etc., portable digital devices for intelligent mobile com-munication have been very popular. People in daily life can almost get access to the In-ternet anytime and anywhere; represented by WeiBo, DouBan, TianYa, ZhiHu, WeChat, etc., social media platforms promote the communication between people, people are free to share their mood and ideas on a social platform; represented by TaoBao, groupon, WeiShang, etc., the emergence of new electronic business models promote the prosperity of electronic commerce. All these phenomena should owe to the rapid development of IT. The combination of IT with fields such as financial and entertainment industries has driven us enter the era of Big Data. Big Data not only challenges the rapid and healthy development of human society, but also provides people from all walks of life with un-precedented opportunities.Many researchers have pointed out that the scientific problems that Big Data faces may essentially be that of complex networks, and network science should be the corner-stone of Big Data technology and industry. Structure analytics of complex networks is the basis of network science. The study of network structure can help to understand and predict the functions and behaviors of complex networks. The community structure is one of the most basic and important structure features of complex networks, and complex net-work community structure analysis is an important theoretical basis of the analysis of other network characteristics. Due to the fact that, in reality a lot of networks are dynamically evolving over the time, consequently, to study the balance structure of complex networks may provide theoretical support to the research on network dynamical evolution.This dissertation is mainly focused on the community structure mining from complex networks and the network balance structure transformation problem and the adopted re-search method is based on swarm intelligence optimization. The idea is to model network issues as optimization problems and then design efficient swarm intelligence optimiza-tion based algorithms considering the structural characteristics of the networks so as to solve the modeled optimization problems. Swarm intelligence optimization is a class of bionic optimization algorithms, and it is characterized by the implicit parallelism which makes it suitable for solving large-scale network optimization problem. Among swarm intelligence optimization algorithms, particle swarm optimization algorithm is outstand-ing for its simple principle, easy implementation and less parameters. This paper mainly studies particle swarm optimization based complex network structure analysis problems. The main research contents include:1. This dissertation briefly lists some important issues concerning complex network structure analysis, with emphasis on the community structure and the balance struc-ture. The definitions of network community structure and balance structure are given. This dissertation introduces the basic ideas of evolutionary computation and swarm intelligence optimization. It also gives introduction to evolutionary multi-objective optimization and some related concepts. It presents the rationale of canon-ical particle swarm optimization algorithm, and lists some classical particle swarm optimization algorithm for solving multi-objective optimization problems.2. After introducing the related theories of network structure analysis and particle swarm optimization, this paper puts forward an efficient single objective particle swarm optimization algorithm for solving the complex network community mining task. The proposed algorithm maximizes the modularity function so as to find the corresponding network community structure. Since the canonical particle swarm optimization algorithm and its most variants are mainly designed for solving con-tinuous optimization problems, but the optimization of the modularity function is a discrete problem, consequently, in the proposed algorithm we take full considera-tion of the characteristics of the community mining problem, and we redefine the discrete representation of a particle and reformulate the discrete status update equa-tions. Due to the fact that many real networks are large in size, in order to improve the global optimization ability of the particle swarm optimization algorithm, we add a greedy local search strategy to the discrete particle position update equation. To validate the effectiveness of the proposed algorithm, a large number of simula-tion experiments on synthetic and real world network data sets have been carried out, and 7 algorithms existed in the literature are compared against the proposed algorithm, the experiments show that the proposed greedy discrete particle swarm optimization algorithm not only has a good community mining ability, but also can deal with networks with moderate scales in a reasonable amount of time.3. Because the optimization of the modularity function has the resolution limit, i.e., modularity optimization may not identify modules smaller than a certain scale, which depends on the total size of the network and the degree of interconnected-ness of the modules. In order to overcome the resolution limit, after considering the advantage of multi-objective optimization and the definition of network com-munity, a multi-objective optimization model of community mining is established. On the basis of the former proposed discrete particle swarm optimization algorithm, we put forward a decomposition based multi-objective particle swarm optimization algorithm for complex network community mining. In view of that the established multi-objective optimization model is for unsigned network, however, in reality a lot of networks are signed, namely there exist friendly and antagonistic relationships between the network members. In order to make the algorithm capable of handling signed networks, we extend the former proposed optimization model, we modify particles’status update rules based on the special network topology information. In order to verify the validity of the algorithm, a large number of simulation exper-iments on synthetic and real world network data sets have been carried out, and 10 algorithms existed in the literature are compared against the proposed algorithm, the experiments show the effectiveness of the algorithm.4. Community structure is one of the notable features of complex networks, com-munity mining can provide inspiration for the research on other network features. Based on our previous studies on the signed and unsigned network community structure analytics, we do research on the balance structure of social networks. The research on the balance structure of social networks not only may help to study the distribution of the relationship between the network members, but also can assist decision makers to take positive measures to help better communication between network members so as to build a harmonious network. Based on our previous re-search, we put forward an algorithm that can transform a structurally unbalanced so-cial network into a network with balanced structure. The proposed method consists of two steps, the task of the first step is to discover the community structure hidden behind a social network, and the adopted community mining avenue is the multi- objective particle swarm optimization algorithm. The first step of the algorithm can ensure that there are as many friendly relations as possible within a community while hostile relations between communities. The second step is to determine the unbalanced edges from the community structure obtained by the first step, putting it another way, to figure out the friendly relations between communities and hostile relations within a community, afterwards, inverse the attributes of these unbalanced edges. From the perspective of sociology, the costs of the inverse of the attributes of those unbalanced edges are different, because the premium needed to turn two ene-mies into friends is much higher than that of the opposite way. Since the first step of the proposed method can produce a number of different network community struc-tures, so different community partition may lead to different consumption. This paper designs a feasible solution selection strategy, and this strategy can choose the best solution from the output of the first step as the input of the second step. In addition, in order to improve the search ability of the particle swarm optimization algorithm used in the first step, we modify the subproblem update strategy. The effectiveness of the proposed method is verified on a large number of network data sets.