Graphs and networks are a useful abstraction to represent a wide range of systems. Sets of nodes that are more highly interconnected constitute a ‘community’. Community detection algorithms help to reveal a decomposition of a network in modules. These communities can overlap, and nodes can have several community memberships. We present M-Link, a memetic algorithm for overlapping community detection. It maximises an objective function called link partition density. The communities of edges obtained with this method naturally translate to overlapping communities of nodes. The method is based on local expansion and a specialised local search mechanism. Label propagation methods are used for initialising a multi-agent tertiary tree population structure. We use the normalised mutual information to evaluate the similarity between the known community structure in a collection of benchmark networks and the community structure detected by M-Link. The method outperforms other state-of-the-art algorithms for overlapping community detection and it has better accuracy and stability.

中文翻译：

---

M-Link：用于重叠社区检测的链接聚类模因算法

图形和网络是表示广泛系统的有用抽象。高度互连的节点集构成一个“社区”。社区检测算法有助于揭示模块中网络的分解情况。这些社区可以重叠，并且节点可以具有多个社区成员身份。我们提出了M-Link，一种用于重叠社区检测的模因算法。它最大化了称为链接分配密度的目标函数。用这种方法获得的边社区自然转化为节点的交迭社区。该方法基于本地扩展和专门的本地搜索机制。标签传播方法用于初始化多主体三级树种群结构。我们使用标准化的互信息来评估基准网络集合中的已知社区结构与M-Link检测到的社区结构之间的相似性。该方法优于其他用于重叠社区检测的最新算法，并且具有更好的准确性和稳定性。