一种基于密度的无监督联系发现方法  

A New Method of Unsupervised Link Discovery Based on the Relative Density

作　　者：吴姗[1,2] 倪志伟[1,2] 罗贺[2,3] 郑盈盈[1,2] 

机构地区：[1]合肥工业大学管理学院,安徽合肥230009 [2]过程优化与智能决策教育部重点实验室,安徽合肥230009 [3]合肥工业大学计算机网络系统研究所,安徽合肥230009

出　　处：《中国管理科学》2008年第S1期29-32,共4页Chinese Journal of Management Science

基　　金：国家高科技研究发展计划(863)资助项目(2007AA04Z116)

摘　　要：在数据挖掘过程中,利用K-近邻(KNN)算法搜索新颖节点往往具有一定的局限性和偏差性。针对此问题,本文提出了加权距离和相对密度的概念,采用基于加权距离的相对密度来度量一个对象的局部离群程度。在此基础上,提出了一种基于密度的无监督联系发现方法,并进行了实验。实验结果表明,该方法能够较准确地描述对象的异常程度,具有更高的精确度。Using K-neighbor(KNN)algorithm to solve novel node discovery problems usually has certain limitations and deviations during the process of data mining.The paper presents the concepts of the weighted distance and the relative density according to the above problems,and measures the local outlier degree of an object by its relative density based on the weighted distance.On this basis,the paper suggests a new method of unsupervised link discovery based on the relative density.The experiment shows that the new method has a better precision and can describe the outlier degree of the object more accurately.

关 键 词：无监督联系发现 新颖节点发现 加权距离 相对密度 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]