基于半监督聚类的局部网络拓扑测量任务选取方法  

A method for selecting topology measurement task of local network based on semi-supervised clustering

作　　者：张晔 张宇[1] ZHANG Ye;ZHANG Yu(School of Computer Science and Technology,Harbin Institute of Technology,Harbin 150001,China)

机构地区：[1]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《智能计算机与应用》2018年第5期89-94,99,共7页Intelligent Computer and Applications

基　　金：国家重点研发计划(2016YFB0801303-2)

摘　　要：本文提出了一种基于半监督聚类的测量任务选取方案以提高局部网络对外连接IP地址的发现效率。首先选择少量测量任务执行测量作为标记样本,计算已知类别的质心,然后利用未标记样本到最近质心的距离进行聚类,最后从距离已知类别较远的未标记样本中生成新的测量任务,迭代执行直到未发现新的类别。半监督聚类中相关参数用控制变量法进行选取。通过发现的局部网络对外连接IP地址数量分析测量效率,并利用聚类的外部指标评价本文算法的聚合能力。This paper proposes a selection scheme of measurement task based on semi-supervised clustering,aiming at improving the discovery efficiency on IP addresses in local network that connected to external network. Three main approaches are conducted.Firstly,a small part of measurement tasks is selected to perform measurements as marker samples,and calculate centroids of all the known classes. Then,clustering is implemented using the distances from unlabeled samples to centroids. At last,measurement tasks are generated from the unlabeled samples that are far from all the centroids,by iterating this way until no any newclass can be found. The parameters of the semi-supervised clustering are determined by the control variable. The measurement efficiency of the tasks is analyzed via the obtained number of IP addresses in local network that connected to external network. Furthermore,the aggregation ability of overall algorithm can be evaluated using clustering external indicator.

关 键 词：拓扑测量 半监督聚类 局部网络对外连接IP地址 

分 类 号：TP393.4[自动化与计算机技术—计算机应用技术]