Research For Community Detection Algorithms In Complex Network

Community structure commonly exists in complex network as a character of network,mainly presented groups such that the connections within each group are dense, whileconnections between groups are sparse. The community structure of network showshierarchical characteristics mainly because large groups contain small communities andsmaller ones embody more small-scale parts. The hierarchical division algorithms arecontinuously proposed to detect community structure, reaching a new level whenmodularity is raised by Newman. However, with study of network developing, facing thescale of network increasing ceaselessly, multiple algorithms are hard to adapt to explorecommunity in large-scale network, appearing out the deficiencies of low accuracy ofalgorithm partitioning. Aiming at the precision problem of complex network algorithm andhow to gain the better consequence in weighted networks when detected, this paper doesthe work as follows:Firstly, with regard to low accuracy of algorithms for detecting community structure,proposed an algorithm for finding community structure in unweighted networks based onnode dissimilarity, algorithm selects core nodes by the standard of degree and closeness,forming first division, adopting the global optimization strategy to make division again,consequently achieving a level division of the network effectively. Experimental resultsshow the accuracy the algorithm.Secondly, as to weighted network, constructing weighted stock network based onactual stock data. Because of differences of the role in stock network, firstly selectingactive nodes by defining the activeness of stocks, then making the first merge bycommunity dissimilarity, forming the highly cohesive small-scale community structure,secondly making division based on small-scale community structure. By analysis of theexperimental results, the algorithm can effectively elect active nodes of network, andmaking comparisons with the consequences when other algorithms applied to establish theweighted network highlights the superiority of the algorithm.