Cai Lei, Zhou Jincheng, Wang Dan
Key Laboratory of Complex Systems and Intelligent Optimization of Guizhou Province, School of Computer and Information, Qiannan Normal University for Nationalities, Duyun, Guizhou, China.
State Key Laboratory of Public Big Data, College of Computer Science and Technology, Guizhou University, Guiyang, China.
PeerJ Comput Sci. 2024 Jul 16;10:e2185. doi: 10.7717/peerj-cs.2185. eCollection 2024.
As one of the essential topological structures in complex networks, community structure has significant theoretical and application value and has attracted the attention of researchers in many fields. In a social network, individuals may belong to different communities simultaneously, such as a workgroup and a hobby group. Therefore, overlapping community discovery can help us understand and model the network structure of these multiple relationships more accurately. This article proposes a two-stage multi-objective evolutionary algorithm for overlapping community discovery problem. First, using the initialization method to divide the central node based on node degree, combined with the cross-mutation evolution strategy of the genome matrix, the first stage of non-overlapping community division is completed on the decomposition-based multi-objective optimization framework. Then, based on the result set of the first stage, appropriate nodes are selected from each individual's community as the central node of the initial population in the second stage, and the fuzzy threshold is optimized through the fuzzy clustering method based on evolutionary calculation and the feedback model, to find reasonable overlapping nodes. Finally, tests are conducted on synthetic datasets and real datasets. The statistical results demonstrate that compared with other representative algorithms, this algorithm performs optimally on test instances and has better results.
作为复杂网络中重要的拓扑结构之一，社区结构具有重要的理论和应用价值，吸引了众多领域研究人员的关注。在社交网络中，个体可能同时属于不同的社区，如工作小组和兴趣小组。因此，重叠社区发现有助于我们更准确地理解和建模这些多重关系的网络结构。本文针对重叠社区发现问题提出了一种两阶段多目标进化算法。首先，利用基于节点度的初始化方法划分中心节点，结合基因组矩阵的交叉变异进化策略，在基于分解的多目标优化框架上完成第一阶段的非重叠社区划分。然后，基于第一阶段的结果集，从每个个体社区中选择合适的节点作为第二阶段初始种群的中心节点，并通过基于进化计算和反馈模型的模糊聚类方法优化模糊阈值，以找到合理的重叠节点。最后，在合成数据集和真实数据集上进行测试。统计结果表明，与其他代表性算法相比，该算法在测试实例上表现最优，具有更好的效果。
