基于群体智能的自组织重叠社团结构分析算法

社团结构分析是复杂网络研究的一项重要内容。基于群体智能思想提出了一种自组织的重叠社团结构分析算法SO^2CSA^2。基本思想是：把网络视为一个群体，网络节点是其中的一个个具有简单智能的个体，每个个体依据定义的社团连接分数自主决定要加入的社团（可同时加入多个社团）。首先在网络中寻找一组K-派系作为初始社团结构；在此基础上，所有个体迭代地选择其社团归属，最终整个网络的社团结构将逐渐生长出来；最后对获得的社团结构进行后处理，即调整少量节点的社团归属，以提高其质量。在一组合成网络和现实世界网络上的实验表明，SO^2CSA^2发现的社团结构的质量比两种对比算法（SLPA和OSLOM）更好，尤其是在网络中重叠节点较多或节点重叠度较大的情况下，社团结构质量的提升更为明显。

Self-organizing overlapping community structure analysis algorithm based on swarm intelligence

Community structure analysis is a critical task in examining a complex network. This paper presented a self-organizing overlapping community structure analysis algorithm (SO^2CSA^2) based on the swarm intelligence theory. The basic idea behind the algorithm was that it treats an analyzed network as a swarm intelligence system, of which each node was an individual with simple intelligence. Each individual independently decides to which community it joined based on a defined metric named connection score. An individual could join to multiple communities simultaneously. At first, the algorithm found a set of K-cliques from the analyzed network as the initial community structure. Then, each individual in the system acted iteratively to join into or leave from communities, and an optimal community structure of the whole network could develop and eventually emerge from the initial community structure. Finally, to improve the quality of the obtained community structure, a post process adjusted community assignments of a small number of individuals. Experimental evaluation on a number of synthesized networks and real-world networks indicates that the quality of community structures discovered by SO^2CSA^2 is better than those of two compared algorithms, SLPA and OSLOM, especially on networks with a large number of overlapping nodes or on networks with overlapping nodes of which overlapping degrees are high.