基于贡献函数的重叠社区划分算法  

Overlapping-communities Recognition Algorithm Based on Contribution Function

作　　者：刘功申[1] 孟魁[1] 郭弘毅 苏波[1] 李建华[1] 

机构地区：[1]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《电子与信息学报》2017年第8期1964-1971,共8页Journal of Electronics & Information Technology

基　　金：国家973关键技术研究项目(2013CB329603);国家自然科学基金(61472248)~~

摘　　要：现实世界中的网络结构呈现出重叠社区的特征。在研究经典的标签算法的基础上,该文提出基于贡献函数的重叠社区发现算法。算法将每个节点用三元组(阈值、标签、从属系数)集合来表示。节点的阈值是每次迭代过程中标签淘汰的依据,该值由多元线性方程自动计算而来。从属系数用于衡量当前节点与标签所标识社区的相关度,从属系数的值越大说明该节点与标签所标识社区的关联性越强。在每一次迭代的过程中,算法依据贡献函数计算每个节点的从属系数,并生成新的三元组集合。然后依据标签决策规则淘汰标签,进行从属系数规范化。通过对真实的复杂网络和LFR(Lancichinetti Fortunato Radicchi)自动生成的网络进行测试可知,该算法的社区划分准确率高,而且划分结果稳定。Overlapping is one of the most important characteristics of real-world networks. Based on the classic labeling algorithm, the overlapping-community orientated label propagation algorithm based on contribution function is proposed. In this algorithm, each node is indicated by a set of triples （threshold, label, and coefficient）. The threshold value of every node is used as a metric for labels decision, which is calculated automatically by multiple linear regression equation. The dependent coefficient is used to measure the relevance of the current node with the correspondent community which is marked by the label. A greater value of dependent coefficient means a stronger association between the node and the community. During each iteration process, the dependent coefficients are calculated through Contribution Function （CF） of each node, and new triples are produced. Then the labels in terms of decision rules are selected, and the dependent coefficients of the node are normalized. According to the tests with real-world networks and automatic generation of LFR （Lancichinetti Fortunato Radicchi） test network, the algorithm can divide communication with high accuracy and robust result.

关 键 词：复杂网络 社区发现 重叠社区 

分 类 号：TP309[自动化与计算机技术—计算机系统结构]