第 39 卷第 9 期 电 子 与 信 息 学 报 Vol.39No.9
2017 年 9 月 Journal of Electronics & Information Technology Sept. 2017
基于矢量影响力聚类系数的高效有向网络社团划分算法
邓小龙
*①
翟佳羽
②
尹栾玉
③
①
(北京邮电大学网络空间安全学院可信分布式计算与服务教育部重点实验室 北京 100876)
②
(北京邮电大学国际学院 北京 100876)
③
(北京师范大学中国社会管理研究院 北京 100875)
摘 要:社团结构划分对于分析复杂网络的统计特性非常重要,以往研究往往侧重对无向网络的社团结构挖掘,对
新兴的微信朋友圈网络、微博关注网络等涉及较少,并且缺乏高效的划分工具。为解决传统社团划分算法在大规模
有向社交网络上无精确划分模拟模型,算法运行效率低,精度偏差大的问题。该文从构成社团结构最基础的三角形
极大团展开数学推导,对网络节点的局部信息传递过程进行建模,并引入概率图有向矢量计算理论,对有向社交网
络中具有较大信息传递增益的节点从数学基础创造性地构建了有向传递增益系数(Information Transfer Gain,
ITG)。该文以此构建了新的有向社团结构划分效果的目标函数,提出了新型有向网络社团划分算法 ITG,通过在
模拟网络数据集和真实网络数据集上进行实验,验证了所提算法的精确性和新颖性,并优于 FastGN, OSLOM 和
Infomap 等经典算法。
关键词:有向社团划分;信息传递增益;目标函数优化;算法可扩展性
中图分类号: TP393; TP391
文献标识码: A 文章编号:1009-5896(2017)09-2071-10
DOI: 10.11999/JEIT170102
Vector Influence Clustering Coefficient Based Efficient
Directed Community Detection Algorithm
DENG Xiaolong
①
ZHAI Jiayu
②
YIN Luanyu
③
①
(Key Laboratory of Trustworthy Distributed Computing and Service of Education Ministry, Beijing University of Posts and
Telecommunications, Beijing 100876, China)
②
(International School, Beijing University of Posts and Telecommunications, Beijing 100876, China)
③
(China Academy of Social Management, Beijing Normal University, Beijing 100875, China)
Abstract: Community detection method is significant to character statistics of complex network. Community
detection in directed structured network is an attractive research problem while most previous approaches attempt
to divide undirected networks into communities while there has appeared many large scale directed social network
such as WeChat circle of friends and Sina Micro-Blog. To solve the problem that low quality of model, low efficiency
of execution and high deviation of precision from the conventional community detection algorithm on large-scale
social network and directed network, this paper provides an approach that starts with the triangle structure of
community basis and models the local information transfer to detect community in large-scale directed social
network. Basing on the directed vector theory in probability graph and the high information transfer gain of vertex
in directed network, this paper constructs the Information Transfer Gain (ITG) method and the corresponding
target functions for evaluating the quality of a specific partition in community detection algorithm. Then the
combine of ITG with the target function to compose the new community detection algorithm for directed network.
Extensive experiments in synthetic signed network and real-life large networks derived from online social media, it
is proved that the proposed method is more accurate and faster than several traditional community detection
methods such as FastGN, OSLOM and Infomap.
Key words: Community detection in directed network; Information Transfer Gain (ITG); Target function
optimization; Scalability