| In recent years,research on complex networks has been widespread public concern.In real life,there are many complex systems which are transformed into corresponding complex networks through mathematical modeling.The community structure represents different division status of nodes in the complex network.It is very meaningful to dig the community structure information in the network.It can help us analyze the relationships between things in the network,so as to clearly grasp the overall structure and characteristics of the network,and tap into the internal potential patterns of the network to understand its internal structure and characteristics,even forecast the network behavior.However,most of the algorithms for solving the problem of community division of complex networks still have certain flaws at present.For example,it is necessary to know the number of community divisions in advance,or some thresholds and so on.There is also a problem that the time complexity is too high,which does not apply to large-scale complex network division problems.MODTLBO/D algorithm is a teaching-learning-based multi-objective optimization algorithm.It is a new intelligent optimization algorithm proposed by Chen in 2016.It is the first time to adopt the teaching-learning-based optimization strategy to solve the problem of community detection.Through the teaching phase(students get knowledge from the teacher)and the learning phase(students learn from each other),we can improve students’ grades together.In this algorithm,each individual learns from the average value of neighbors.The time complexity of the algorithm is high.In the learning phase,individuals learn from their neighbors and it is easy to fall into the local optimum.In order to improve the multi-objective optimization of community detection using discrete teaching-learning-based optimization with decomposition,called MODTLBO/D,and decrease time complexity,we propose an efficient teaching-learning-based optimization algorithm combined with multi-population evolutionary strategy for community detection.In our paper,we adopt adaptive learning factor in teacher phase to enhance the ability of exploration and search.In learner phase,each learner employs the random learning strategy or modified quantum-behaved learning strategy in corresponding subpopulation.Individuals are preprocessed when individuals are initialized.After each generation,subpopulations exchange information to maintain the diversity and discourage premature convergence.By comparing a large number of experiments,E-MODTLBO/D is superior to MODTLBO/D and other classic community detection algorithms in terms of time complexity and discovery of high-quality community structure.This algorithm is more competitive than other current algorithms. |
