基于节点拓扑结构和属性的重叠社区检测算法

近些年来,网络中的重叠社区检测应用越来越广泛,这极大引起了人们的兴趣。通常重叠社区检测只考虑节点的拓扑结构信息,忽略了节点的属性信息,这可能会遗漏数据间的重要结构。本文提出了一种基于节点拓扑结构和属性相似度的局部社区检测算法。首先,计算出节点与社区之间属性相似度。其次,为挖掘内部紧耦合的社区,计算出能够表示耦合性的模块度增量。最后,通过融合多个已检测到局部社区,计算出隶属矩阵从而获取全局重叠社区结构。在三个有真实社区的数据集上的实验结果从内部指标模块度和与真实社区的比较验证了基于节点拓扑结构和属性相似度的重叠社区检测算法比目前基于节点结构的社区检测算法在效率和有效性上有所提升。

Overlapping Community Detection With Node Structure and Attribute

In recent years, the application of overlapping community detecting of complex networks becomes more widely, which aroused great interest. Usually overlapping community detection only considers the topology information of nodes, and ignores the attribute information, which may omit the important structure of the data. This paper proposed a Local Community Detection (LCD) algorithm which is based on the topology and the similarity of node attribute to find an overlapping community from a seed vertex. First, calculated the similarity between the node attribute and the community. Second, in order to mine the tight coupling of the internal community, calculated the increment of modularity which represents the coupling. At the end, merged multiple local communities which have been detected and calculate a membership matrix, which can been seen as the global overlapping community structure of a graph. On three data sets with ground truth communities, experimental results from the internal index module and the comparison with the ground truth community verify that the algorithm based on node topology and attribute similarity can be improved in the efficiency and effectiveness of community detection algorithm based on node structure.