模块密度谱分的网络社团发现方法

为有效地检测复杂网络中的社团结构，对评估与发现社团的模块密度函数(即D值)进行了优化．通过模块密度函数的优化进程，论证了模块密度函数被优化框定到广阔的谱分聚类方法中的矩阵松散最大化，并且提出了一种新的谱分算法．该算法允许自动选择最优的社团结构数目．在经典的计算机产生的随机网络及真实世界网络中检验了该算法．特别地，当网络中社团结构变得模糊时，实验结果显示这种新的算法在发现复杂网络社团上比基于模块密度的直接核方法及基于模块函数(Q)的谱分方法更加有效．

Spectral approach to finding communities in networks based on the modularity density

To detect the community structure in complex networks effectively, the modularity density function (D value) is optimized by the optimizing process, how the optimization of the D function can be reformulated as a spectral relaxation problem is proved and a new spectral clustering algorithm is proposed. The algorithm allows automatic selection of the number of community structures. The approach is illustrated and compared with the direct kernel approach based on the modularity density and spectral clustering based on modularity (Q) by using a classic computer generated networks and a real world network. Experimental results show the significance of the proposed approach, particularly, in the cases when the community structure is obscure.