基于节点拓扑结构和属性的重叠社区检测算法

发布时间：2018-01-01 21:37

本文关键词：基于节点拓扑结构和属性的重叠社区检测算法　出处：《计算机应用研究》2016年12期 　论文类型：期刊论文

【摘要】：针对已有重叠社区检测通常只考虑节点的拓扑结构信息,忽略了节点的属性信息,导致数据间的重要结构遗漏的问题,提出了一种基于节点拓扑结构和属性相似度的重叠社区检测算法。首先,基于余弦相似度计算候选节点和局部社区之间的相似度,提高局部搜索效率;其次,改进局部模块度增量计算方法,使局部搜索模型收敛于发现潜在的真实社区;通过融合多个已检测到的局部社区计算隶属矩阵,从而获取全局重叠社区结构;最后,在真实数据集上,与已有基于拓扑结构的社区检测算法进行实验对比。结论表明,该算法在模块度和F1-measure的指标上取得了较好的表现且更适用于稀疏网络。
[Abstract]:In view of the existing overlapping community detection, only the topology information of nodes is usually considered, and the attribute information of nodes is ignored, which leads to the omission of important structures between data. An overlapping community detection algorithm based on node topology and attribute similarity is proposed. Firstly, the similarity between candidate nodes and local communities is calculated based on cosine similarity to improve the local search efficiency. Secondly, the method of local modular increment is improved to make the local search model converge to discover the potential real community. The global overlapping community structure is obtained by merging multiple detected local community computing membership matrices. Finally, in the real data set, compared with the existing community detection algorithm based on topology, the conclusion shows that. The algorithm achieves better performance in modularity and F1-measure and is more suitable for sparse networks.
【作者单位】：南京财经大学信息工程学院;
【分类号】：TP301.6
【正文快照】： 0引言网络通常由相互连接的动态节点组成。社交网络、生物网络和计算机科学网络仅仅是复杂网络的一些代表,并且这些网络通常展示了具有拓扑特性的社区。社区或模块结构被认为具有现实社交网络的重要特性。从网络中发现这些潜在的社区可视为将节点集合聚类成社区的问题。常见的

【相似文献】