结合网络拓扑与节点内容的统一化半监督社团检测方法  

A Unified Semi-supervised Community Detection Approach Integrating Network Topology and Node Contents

作　　者：许伟忠 曹金鑫 金弟 孙翔 张晓峰[1] 刘路 丁卫平[1] Xu Weizhong;Cao Jinxin;Jin Di;Sun Xiang;Zhang Xiaofeng;Liu Lu;Ding Weiping(School of Information Science and Technology,Nantong University,Nantong 226019,China;College of Intelligence and Computing,Tianjin University,Tianjin 300350,China;School of Informatics,University of Leicester,Leicester,UK LE17RH)

机构地区：[1]南通大学信息科学技术学院,江苏南通226019 [2]天津大学智能与计算学部,天津300350 [3]莱斯特大学信息学院莱斯特,英国LE17RH

出　　处：《南京师大学报（自然科学版）》2023年第1期130-138,共9页Journal of Nanjing Normal University(Natural Science Edition)

基　　金：国家自然科学基金面上项目(61976120);江苏省自然科学基金面上项目(BK20191445);江苏省高等学校自然科学研究面上项目(21KJB520018);南通大学人才引进项目(03081198)。

摘　　要：在复杂网络分析中,社团检测发挥着越来越重要的作用,而在实际应用中如何提高社团检测的性能仍是一个共同研究目标.由于网络节点中内容信息有助于社团识别,一些方法侧重于将网络拓扑和节点内容相结合,并且获得了不错效果.此外,也有些方法借用节点之间的拓扑相似度,以提升实现社团检测性能.鉴于此,我们提出了一个统一化方法,结合节点内容的半监督社团检测,简称SCDNC.在该方法中,我们不仅将链接增强应用于社团检测,而且实现了拓扑和内容有机融合.首先,我们运用随机模型来描述节点社团隶属度.其次,我们构建出一个刻画节点内容社团隶属度的随机块模型,节点社团隶属度作为节点内容的权重向量,以实现拓扑和内容结合.再次,我们利用网络中节点之间的拓扑相似度构建先验信息,即,使网络中节点与其最相似的邻居节点具有相同的隶属度分布.最后,使用非负矩阵分解的方法学习新模型的统一化参数.在带有真实标签的人工网络和真实网络上,我们对新方法与一些当前流行的社团检测方法进行了性能比较.实验结果显示,通过融合节点内容和先验信息强化的链接,新方法检测社团的性能取得了显著提升.Community detection plays an increasing important role in complex network analysis.There is still a goal that how to improve the performance of community detection in real applications.Due to the content in networks helpful to identifying communities,some methods focus on combining network topology with node content,which obtains no bad performance of community detection.Besides,some community detection enhancement methods are mainly based on designing the topological similarity of nodes to adjust network topology,which aims to achieve the enhancement.In order to further improve the quality of community detection,we propose a unified method,Semi-supervised Community Detection with Node Contents,shorted as SCDNC,which not only apply the enhancement into community detection,but also achieve the integration of network topology and node content.In the new method,firstly,we propose a stochastic block model to describe the community memberships of nodes.Secondly,we present another stochastic model to describe the community memberships of node contents,which utilizes community memberships of nodes as weight vectors of node contents.By now,integrating network topology with node content is achieved.Thirdly,we calculate the topological similarity of nodes by using links,and then model the prior information based on topological similarity,i.e.,we make nodes and their most similar neighbors have the same community membership.Finally,we present a nonnegative matrix factorization approach to obtain the parameters of the model.On both synthetic and real-world networks with ground-truths,we compare performance of the new method with the state-of-the-art methods.The experimental results show that the new method obtains significant improvement for community detection via combining node contents and network topology enhanced by prior information.

关 键 词：社团检测 节点内容 先验信息 随机块 非负矩阵分解 

分 类 号：TP182[自动化与计算机技术—控制理论与控制工程]