Page 408 - 《软件学报》2025年第5期
P. 408
软件学报 ISSN 1000-9825, CODEN RUXUEW E-mail: jos@iscas.ac.cn
2025,36(5):2308−2320 [doi: 10.13328/j.cnki.jos.007197] [CSTR: 32375.14.jos.007197] http://www.jos.org.cn
©中国科学院软件研究所版权所有. Tel: +86-10-62562563
*
面向属性网络社团检测的度修正广义随机块模型
王 笑, 戴 芳, 郭文艳, 王军锋
(西安理工大学 理学院, 陕西 西安 710054)
通信作者: 戴芳, E-mail: daifang@xaut.edu.cn
摘 要: 随机块模型可以拟合各种网络的生成, 挖掘网络的隐含结构与潜在联系, 在社团检测中具有明显的优势.
广义随机块模型 GSB 是基于链接社团的思想发现广义社团的, 但其仅适用于有向无属性网络. 针对无向属性网络,
对网络拓扑信息建模的同时对节点属性进行建模, 提出一种度修正的属性网络广义随机块模型 DCGSB (degree
corrected general stochastic block model). 在该模型中, 假设网络拓扑信息和属性信息的生成过程都服从幂函数形式
的分布, 并且引入节点的度来刻画网络的无标度特性, 可以更好地拟合真实网络的生成. 利用期望最大化算法对
DCGSB 模型的参数进行估计, 通过硬划分处理, 得到节点隶属度, 进而完成社团检测. 在 3 个含有不同结构的真实
属性网络数据集上进行实验, 并与 10 种社团检测算法进行对比, 结果表明 DCGSB 模型不仅继承了 GSB 模型的优
点, 能发现广义社团, 而且由于属性信息和节点度的引入, 使其社团检测能力优于其他 10 种比较算法.
关键词: 随机块模型; 节点度; 属性网络; 社团检测
中图法分类号: TP311
中文引用格式: 王笑, 戴芳, 郭文艳, 王军锋. 面向属性网络社团检测的度修正广义随机块模型. 软件学报, 2025, 36(5): 2308–2320.
http://www.jos.org.cn/1000-9825/7197.htm
英文引用格式: Wang X, Dai F, Guo WY, Wang JF. Degree Corrected General Stochastic Block Model for Community Detection in
Attributed Network. Ruan Jian Xue Bao/Journal of Software, 2025, 36(5): 2308–2320 (in Chinese). http://www.jos.org.cn/1000-9825/
7197.htm
Degree Corrected General Stochastic Block Model for Community Detection in Attributed Network
WANG Xiao, DAI Fang, GUO Wen-Yan, WANG Jun-Feng
(School of Science, Xi’an University of Technology, Xi’an 710054, China)
Abstract: Stochastic block models can fit the generation of various networks, mining implicit structures and potential connections within
these networks. Thus, they have significant advantages in community detection. General stochastic block (GSB) models discover general
communities based on link communities, but they are only applicable to directed non-attributed networks. This study proposes a degree
corrected general stochastic block (DCGSB) model for undirected attributed networks which models both network topology information
and node attributes. In the DCGSB model, it is assumed that the generation of network topology information and node attributes follows a
distribution in the form of power functions. Node degrees are introduced to characterize the scale-free property of networks, which allows
the model to better fit the generation of real networks. The expectation-maximization algorithm is employed to estimate the parameters of
the DCGSB model, and node-community memberships are obtained by hard partition to complete community detection. Experiments are
conducted on three real attributed network datasets containing different network structures, and the proposed model is compared with ten
existing community detection algorithms. Results show that the DCGSB model not only inherits the advantages of GSB models in
identifying general communities but also outperforms the ten algorithms in community detection due to the introduction of attribute
information and node degrees.
Key words: stochastic block model (SBM); node degree; attributed network; community detection
* 收稿时间: 2023-09-23; 修改时间: 2024-01-12; 采用时间: 2024-03-28; jos 在线出版时间: 2024-12-11
CNKI 网络首发时间: 2024-12-12