with dense connections within groups and only sparser connections between them" (Newman, 2004). Such groups are called “communities".
一、LPA算法
当前所用的一些社区发现算法,需要先验信息,比如社团的数目和大小,或者计算量很大。LPA算法只需要使用网络结构,不需要一个预定义目标函数的优化或者先验信息。在LPA中,每一个节点用一个唯一的label初始化,每一步中,每个节点采纳它的多数邻居当前拥有的label,在这个迭代过程中,链接紧密的节点拥有一致的label,从而组成一个community。
二、TCLAP算法(Transitive Clustering and Pruning, TCLAP)
1. Clusters:BFS sampling
随机抽取一个节点v0,定义一个集合C,按照BFS方法增加v0的邻居到集合C。
2. Pruning:
二、realization
1.Newman kindly provided his modularity spectralmaximization algorithm in C,
2.our T-CLAP algorithm is implemented in Python.
3.modified T-CLAP: simpler because it skips clustering
<Peer Influence in the Diffusion of the iPhone3G over a Large Social Network>
三、comparison
T-CLAP and the Newman algorithm, we generally onlycompare the running time, because Newman's algorithm is a graph partitioningalgorithm and does not return any subpopulations (beyond the clus-ters in thereturned partition). For the comparison between T-CLAP and the Clausetalgorithm, the comparison is based on the quality of the subpopulationsreturned, which is measured by the I -E ratio and density, plus speed, which ismeasured by running times.
modularity Q用于评估社团发现的效果,
LPA:Near linear time algorithm to detect community structures in large-scale networks