基于谱聚类的二分网络社区发现算法

Detecting Community from Bipartite Network Based on Spectral Clustering

二分网络是一类特殊的网络,在探索网络深层结构上具有重要作用。针对二分网络社区划分方法仍存在划分精度不高的问题,应用标准化谱聚类,提出了二分网络社区发现算法——谱聚类交互算法(SPCI)。首先,根据二分网络中两类节点之间的连边关系,构建相似性矩阵;然后,利用谱聚类算法将其中一类节点聚类;最后,利用交互度指标实现二分网络的社区划分。在人工数据和真实数据上的验证表明,SPCI不仅拥有比资源分布矩阵算法、边集聚系数算法和联合谱聚类算法更高的准确性和模块度,而且可以较为准确地确定社区划分个数。

Bipartite network is a special kind of network,which plays an important role in exploring the deep structure of the network.However,the methods of dividing the bipartite network community still have some problems,such as low precision of division.Through the application of normalized spectral clustering algorithm,an algorithm of detecting community- spectral clustering interaction (SPCI) was proposed.First,a similarity matrix is constructed based on the relationship between two kinds of nodes.Then,a cluster is clustered by spectral clustering algorithm.Finally,the community partition of two points network is realized by using two kinds of node’s interaction index.Through the verification on artificial data and real data,the result shows that SPCI not only has higher accuracy and modularity than the algorithm based on resource distribution matrix,edge clustering coefficient and spectral co-clustering,but also can accurately determine the number of community partition.