Research On Community Detection Algorithms For Complex Networks

In the era of big data,multiform data need to be processed.In the relational data,individuals can be abstracted as nodes,and the associations among individuals are mapped as the edges of nodes,which can be organized into a complex network with scale-free and small-world.In a complex network,the characteristics of clusters are usually implied.Individuals in a community may come from social groups with similar characteristics and the same living background in the network.If we can effectively dig out the hidden community structure in the network,we can find the important information by the internal and external communities.However,the essence of community detection algorithm is to reveal the communities in the network.In real scenarios,such as recommendation systems,the spread of epidemic diseases,etc.are based on community detection.Therefore,community detection algorithms has important research significance and broad application prospects,and is also one of the hot topics in the field of data mining.The networks are complex and isomerism,in which detecting community structure is more challenging.Although many community detection algorithms have been effectively used to mining network community problems,it has always been the research and improvement direction of community detection algorithms to enhance the quality of community structure and improve the accuracy of community result.For the detect community structure of complex network,this paper mainly does the following work:Firstly,in order to obtain a high-quality community structure,a discrete random drift particle swarm optimization modularity(DRDPSO-net)algorithm is proposed.Firstly,the initial value of particle population is obtained based on the similarity of neighbors between nodes,so that the potential community structure can be found.Secondly,in order to make the traditional random drift particle swarm optimization algorithm to obtain ideal results in network,it is redefined as the discrete updating formula of particles.At the same time,the local community information greedily is updated in the process of local search,which improve the local network structure and gradually enhance the global modularity value.The experimental results show that the algorithm is effective in different scale synthetic networks and real networks.The community structure available in different networks is more desirable.Secondly,in the DRDPSO-net experiment,it is found that the maximum optimization modularity may cause the problem of resolution limitation,and the algorithm is less accurate in the real network division.In order to solve this problem,an mDRDRPSO-net algorithm is proposed.Two conflicting objective functions,Kernel k-means(KKM)and Ratio Cut(RC),were used to control the community size and ameliorate modularity resolution in the network.In addition,Pareto sets are gradually updated according to the multi-objective solution strategy.And the target community structure satisfying the requirements can be selected from Pareto sets.Compared with each algorithm,the mDRDPSO-net algorithm can further improve the accuracy of the network community division.Finally,We briefly analyze the community detection in complex network with attribute information.Base on the similarity of topology and attribute,further explore community mining by the aforementioned multi-objective optimization framework.The experimental results show that this method can also be effectively applied to Facebook-Ego network.