摘要
重叠社区发现已成为复杂网络研究的热点内容。传统基于非负矩阵分解的社区发现方法忽视了特征矩阵选择的重要性,通过模块度优化来确定社区数目导致计算开销大和存在模块度分辨率受限制等问题。针对上述问题,提出了一种基于贝叶斯先验的非负矩阵分解社区发现方法。通过引入贝叶斯非负矩阵分解模型,实现了对社区数目的有效迭代求解。为了得到节点与社区的隶属关系,采用线性转换函数思想定义了社区隶属度指数,并通过设定合理的划分阈值得到网络重叠社区结构。在不同规模的计算机生成和真实世界网络上进行了测试,并与典型算法进行比较,实验结果表明了该算法的可行性和有效性。
Overlapping community detection has become a research hotspot of complex network. Traditional community detection approach based on non-negative matrix factorization ignores the importance of the feature matrix choice, and determines the number of communities by modularity optimization, which has high computation cost and suffers resolution limit. An approach to community detection was proposed based on Bayesian non-negative matrix factorization model with prior parameters. The introduced probabilistic model calculated the number of communities iteratively with high efficiency. To get the node-community affiliation, the membership index between nodes and communities was defined using the idea of linear conversion function, and then overlapping community structure was obtained by setting a reasonable partition threshold. The proposed method was tested on different scale of computer-generated and real world networks, and was compared with typical algorithms. Experimental results confirm the feasibility and effectiveness of the proposed method.
出处
《系统仿真学报》
CAS
CSCD
北大核心
2014年第3期643-649,共7页
Journal of System Simulation
基金
国家863项目(2011AA7032030D)
全军军事研究生课题资助(军事学YJS1062)
关键词
复杂网络
重叠社区发现
特征矩阵
贝叶斯非负矩阵分解
隶属度指数
complex networks
overlapping community detection
feature matrix
Bayesian non-negative matrix factorization
membership index
