摘要
动态网络的社区发现是目前复杂网络分析领域的重要研究内容,然而现有动态网络社区发现方法主要针对同质网络,当网络包含多种异质信息时,现有方法不再适用。针对这个问题,提出了一种基于联合矩阵分解的动态异质网络社区发现方法。首先计算动态异质网络中各个快照图的拓扑相似度矩阵和多关系相似度矩阵;其次利用时序联合非负矩阵分解方法,约束各个时刻快照图的社区划分;最后在真实网络数据集上与K-means、Meta Fac算法进行比较实验,提出算法能够充分利用网络的异质信息与拓扑信息,异质网络社区划分精度优于Meta Fac算法,且划分效果更稳定。结果表明,基于联合矩阵分解的动态异质网络社区发现算法可以有效检测出动态异质网络中潜在的社区结构。
Dynamic community detection is an important research field of complex network analysis. However, with the rapid development of social networks, the structure of networks becomes more and more complex. Traditional community detection methods in homogeneous networks can hardly adapt with the demand of heterogeneous networks. In order to deal with this problem, this paper proposed a joint non-negative matrix factorization algorithm for dynamic heterogeneous networks. Firstly, the algorithm calculated the topology similarity and multi-relational similarity in each snapshot. Then it combined the historical information and current information with joint non-negative matrix factorization algorithm to detect communities from dynamic heterogeneous networks. Finally it took comparative experiment on real Web datasets with K-means and MetaFac algorithms. The proposed algorithm took full advantage of heterogeneous information and network topology information, and could get more accuracy and robust than MetaFac algorithm in heterogeneous network community division. The result of experiments demonstrates that the proposed algorithm can detect community in dynamic heterogeneous network effectively.
出处
《计算机应用研究》
CSCD
北大核心
2017年第10期2989-2992,共4页
Application Research of Computers
基金
国家科技支撑计划资助项目(2014BAH30B01)
国家自然科学基金创新群体资助项目(61521003)
国家自然科学基金资助项目(61379151)
关键词
异质网络
动态网络
社区发现
非负矩阵分解
heterogeneous network
dynamic network
community detection
nonnegative matrix factorization
