Research Of The Dynamic Role Assorted Discovery Of Network Community Based On Particle Swarm Optimization

With the development of internet technologies, e-commerce, online shopping, e-pay and immediate message (IM) have become an essential part of people's daily life. Especially the appearance and growth of SNS, it provides network users with a rather safer platform for communication. However, the implementation of good social service needs to find and construct the users'interesting model, therefore, how to find the similar users, how to discover the community structures in complex network have become a new research spot recently.Traditional community discovery looks on social network service with static viewpoints, for instance, the distribution of node-degree, the edge betweeness, and the coefficient of concentration are used to describe the static features of social network, while ignores the motility of users. This paper comes up with the idea of using particle swarm optimization algorithm to mine the community structures, which pay more attention to the guidiance of elite particles, as well as takes the communities'inner members'distinctive attributes into accout to distribute different roles to them. In terms of the key issues in social network analysis, for example, dynamic discovery, role assort for community inner members and analysis for attack tolerance, to name but a few, this paper based on PSO to study the social network analysis and gets some great achievement as below shows.(1) Applied momentum particle swarm algorithm to realize the dynamic discovery of community in social network. By studying the characteristic matrix of social network, the author comes up with the method of using the top k nontrivial eigenvectors get by Capocci algorithm Vp=(vp1,vp2,…,vps)(p=1,…,k) as input and making use of the PSO algorithm to mine communities. Taking the former r nontrivial eigenvalues before the first negative one, and the corresponding r nontrivial eigenvectors, which means it forecasts that the network has r+1communities. In this condition, r must not less than k=m-1. And on that basis, in order to dynamically find the communities during the optimization procedure to put the choosing process of m into the coding structure. The PSO algorithm provides for the implementation of dynamic role assorted community discovery with theoretical basis.(2) Combined the communities'structure characteristic and the inner members' attributes together to analyze and divide communities on both grammar and semantic levels. It firstly accords to the obvious relationship between nodes in the network to roughly mine the network structures, looks for the communities topology structures in the entire social network, and then defines a theme for each community according to the community's inner members' eigenvectors; Secondly, it analyzes the inner nodes'properties and characteristics to optimize the community dividing results. This algorithm takes advantage of users'characteristics to refine the community discovery results, which is good to study and program the system's functional modules and users'interests units. This algorithm is put forward to optimize the design and implementation of SNS scientific paper management system which will be introduced at the end.(3) Inspired by the priority complex and the grow law, the author puts forward a new dynamic community discovery algorithm based on role assorted thoughts which combines with PSO algorithm. According to the guidance of local optimal particle to the normal particles, the author innovatively comes up with the concept of community seed. The community seed dominants the formation of social communities, the rest particles located around the community seed and form a virtual community with intense inter-inter connections and sparse inter-outer connections. The priority complex and the grow law tell us that early appeared nodes have more chances to accumulate links than late appeared nodes, therefore, the author creatively proposes that:assume the node with the highest degree come to the network firstly, it is appropriate to reversely deduce the entire network's formation and evolution mechanism in accordance with the nodes'degree distribution. In the meantime, distributing roles to the community inner members during the community discovery process, and to dynamically mine community structures. In the part of the design and implementation of SNS scientific paper management, the algorithm is used to cluster papers, and then use clusters as the unit to analyze the relationship between different writers. Using this algorithm is good to find the writers' roles and status in different clusters and to provide better customized recommendation.(4) Based on the dynamic role assorted community discovery algorithm (DRA) and traditional community discovery algorithm (G-N) to construct different kinds of protection strategies. This paper introduces the concept of "flexible degradation", analyzes the strategies' flexible ability to tolerate hostile attacks of this two kind algorithms. The customized protection strategy is able to provide special security for important nodes, then to protect the negotiability and the basic functions of the main network structure, that is to say, to construct the "deep defense strategy" to limit the destructive power of hostile attack. This strategy mainly shows the request to flexibly protect the social network; it aims at increasing the network's stability and robustness on the promise of decreasing the system's protection spending. In the next part, this strategy is applied to the construction of SNS scientific paper management system, running results show that this strategy effectively increases the system's attack tolerance and fast recovery ability, which enables the system to flexibly maintain the information's liquidity and safety.(5) A SNS scientific paper management platform is constructed to analyze and test the suggested algorithms. This platform provides a common academic entrance for scientific research people; it is able to search all kinds of decentralized academic information resources in specific fields, to provide with customized design and recommendation services. Running results of the platform show that the suggested algorithms in this paper increase the platform's adaptability and robustness with great efficiency and flexibility.