Research Of Improved Multi-label Propagation Algorithm Used In Overlapping Community Detection

Community structure,one of the important attribute of complex network,has become a key research direction of complex network.There is a great and profound significance on community detection for understanding the structure and characteristic of network.Therefore,mining the community structure rapidly and accurately in complex network is the most important task.Although scholars have proposed many community detection algorithms,these algorithms can only obtain non-overlapping community structures.In general,overlapping community structure in the complex network of real world,so it is hard to obtain the overlapping community structure accurately through non-overlapping community detection algorithm.Hence,the further research should concentrate on overlapping community detection algorithm.The multi-label propagation COPRA algorithm which is based on the label propagation algorithm is proposed to mine the overlapping community structure.Although the multi-label propagation COPRA algorithm inherits the advantages of simple and efficient from the label propagation algorithm,the disadvantages such as strong randomness and poor robustness still remain,aiming at these problems,an improved multi-label propagation algorithm was proposed,the works in this paper can be summarized as follows:1.This paper was devoted to the theoretical research of complex networks,non-overlapping and overlapping community detection algorithms,gave a brief account of complex networks,and introduced the current research status regarding non-overlapping and overlapping community detection algorithms,then analyzed the basic conception and procedures of several non-overlapping and overlapping community detection algorithms,moreover,pointed out the advantages and disadvantages separately.,2.For the problem of COPRA algorithm using random update order,through LeaderRank sorting algorithm ranks all nodes in the network according to the level of importance,then obtained the order of update,which can reduce the unnecessary updates during the label propagation effectively,accordingly,accelerated the convergence speed of the algorithm and improved the efficiency of the algorithm.3.For COPRA algorithm,in consideration of label update strategy's randomness can cause an unstable result finally,and additional input parameters v is required,this paper redesigned the update strategy by using similarity between nodes and dominant label to make the result more stable with the improved label update strategy and no need to input extra parameters.Through experimental analysis on real network and artificial network data set,the experimental result showed that the improved multi-label propagation overlapping community detection algorithm had a better stability,so the quality of community was increased obviouslty.