A community is the basic component structure of complex networks and is important for network analysis. In recent decades, researchers from different fields have witnessed a boom of community detection, and many algorithms were proposed to retrieve disjoint or overlapping communities. In this paper, a unified expansion approach is proposed to obtain two different network partitions, which can provide divisions with higher accuracies and have high scalability in large-scale networks. First, we define the edge intensity to quantify the densities of network edges, a higher edge intensity indicates a more compact pair of nodes. Second, vertices of higher density edges are extracted out and denoted as core nodes, whereas other vertices are treated as margin nodes; finally we apply an expansion strategy to form disjoint communities: closely connected core nodes are combined as disjoint skeleton communities, and margin nodes are gradually attached to the nearest skeleton communities. To detect overlapping communities, extra steps are adopted: potential overlapping nodes are identified from the existing disjoint communities and replicated; and communities that bear replicas are further partitioned into smaller clusters. Because replicas of potential overlapping nodes might remain in different communities, overlapping communities can be acquired. Experimental results on real and synthetic networks illustrate higher accuracy and better performance of our method.

中文翻译：

---

大规模复杂网络中的统一社区检测算法

社区是复杂网络的基本组成结构，对网络分析具有重要意义。近几十年来，来自不同领域的研究人员见证了社区检测的蓬勃发展，并且提出了许多算法来检索不相交或重叠的社区。本文提出了一种统一扩展的方法来获得两个不同的网络分区，在大规模网络中可以提供更高的划分精度和高扩展性。首先，我们定义边缘强度来量化网络边缘的密度，较高的边缘强度表示更紧凑的节点对。其次，将高密度边的顶点提取出来并表示为核心节点，而将其他顶点视为边缘节点；最后，我们应用扩展策略来形成不相交的社区：紧密相连的核心节点组合为不相交的骨架社区，边缘节点逐渐依附于最近的骨架社区。为了检测重叠社区，采取了额外的步骤：从现有的不相交社区中识别潜在的重叠节点并进行复制；承载副本的社区被进一步划分为更小的集群。因为潜在重叠节点的副本可能保留在不同的社区中，所以可以获取重叠社区。真实和合成网络的实验结果说明了我们方法的更高准确性和更好的性能。从现有的不相交社区中识别潜在的重叠节点并进行复制；承载副本的社区被进一步划分为更小的集群。因为潜在重叠节点的副本可能保留在不同的社区中，所以可以获取重叠社区。真实和合成网络的实验结果说明了我们方法的更高准确性和更好的性能。从现有的不相交社区中识别潜在的重叠节点并进行复制；承载副本的社区被进一步划分为更小的集群。因为潜在重叠节点的副本可能保留在不同的社区中，所以可以获取重叠社区。真实和合成网络的实验结果说明了我们方法的更高准确性和更好的性能。