摘要
为了更便捷准确地进行重叠社区检测,考虑从多个非重叠社区结构中推断出重叠社区,提出一种集成重叠社区检测(IOCD)算法。首先,根据基础社区检测(CD)算法的结果为每个顶点生成一个特征向量,通过这些特征以无监督学习的方式检测密集连接的重叠区域,即利用非重叠CD解来提取与每个顶点相关联的隐性特征信息;然后,不断迭代,最大化每个顶点属于其自身社区的可能性。在合成网络和真实社区网络数据集上进行实验,实验结果表明,在3个标准度量下,所提IOCD算法明显优于其他同类算法,几乎不受基础CD算法的影响。
To detect overlapping communities more conveniently and accurately,an integrated overlapping community detection(IOCD)algorithm is proposed,which considers inferring overlapping communities from multiple non-overlapping community structures.Firstly,an eigenvector is generated for each vertex according to the results of the basic community detection(CD)algorithm.These features are used to detect the overlapping regions with dense connections in an unsupervised learning manner.That is to say,non-overlapping CD solutions are used to extract the implicit feature information associated with each vertex.Then,some iterations are made to maximize the possibility that each vertex belongs to its own community.Experiments on synthetic network and real community network datasets show that the proposed IOCD algorithm is superior to other similar algorithms under three standard metrics,and is almost unaffected by the basic CD algorithm.
作者
陈吉成
陈鸿昶
李邵梅
Chen Jicheng;Chen Hongchang;Li Shaomei(China National Digital Switching System Engineering&Technological R&D Center(NDSC),Zhengzhou 450002,China)
出处
《电子技术应用》
2019年第12期96-100,105,共6页
Application of Electronic Technique
基金
国家自然科学基金创新群体资助项目(61521003)
关键词
社区检测
重叠社区
非重叠社区
顶点
合成网络
community detection
overlapping community
non-overlapping community
vertex
synthetic network