Research On Community Mining In Complex Networks Based On Intelligence Algorithm

Complex networks theory is an important method of complexity science, widely favored by interdisciplinary researchers for a long time. Community structure is one of the network structure characteristics, and the network which has obvious community structure has the characteristics of local accumulation identifying the relationship between the whole and segment. The research of community mining can reveal the structural features of complex networks. We can analyze the topology of the network through community mining, and then predict network structure, analyze the influence on network communication, synchronization and control dynamics and so on. So the research of community mining has important theoretical significance and practical value.Many algorithms are employed in community mining, and modularity-based methods is an important branch. At present, Many intelligent optimization algorithms applied in the field of community mining such as genetic algorithm, ant colony algorithm, particle swarm optimization algorithm and so on. Aimed for the problem of slow convergence speed and low accuracy in community detection, this thesis put forward two kinds of improved algorithms. The main research work and achievements are as follows:(1)The improved discrete particle swarm optimization algorithm(IPSO) is put forward. Firstly, The IPSO is based on the character encoding, and utilized label propagation method based on node importance initialization. Secondly, redefining the speed and position update, and then reordering operation after the update location. Finally adding the clonal selection operator in each iteration. Good initialization method can not only guarantee the diversity of population, but also accelerate the algorithm convergence speed. Clonal selection operator improves the local search ability. Simulation experiments show that the improved discrete particle swarm optimization is superior to the initial position, which guarantees the accuracy of fitness and convergence rapidly.(2)The bat algorithm is a new type of swarm intelligence optimization algorithm, with small amount of calculation and fast convergence rate. The standard bat algorithm is prone to be premature, and only used in continuous domain. For community mining application scenario, the adaptive evolution bat algorithm(AEBA) is put forward. The AEBA is based on the character encoding, used xor operation to processing speed. And then the two-way crossover operator is used for implementation of bats global search, local mutation operator is used for local search. Bat mutation probability depends on the speed of the bat, so as to realize the adaptive evolution of the bat. Simulation experiments are conducted in the seven real networks and the standard synthetic network. To verify the performance of the AEBA is compared with other algorithms. Last the visualization results of Karate and Dolphin network are showed. The experimental results show that AEBA has the advantage of fast convergence rate and high fitness value especially in large-scale networks.