传递闭包聚类中的模糊性分析

The Fuzziness Analysis of Transitive Closure Clustering

传递闭包聚类是根据其相似矩阵的传递闭包生成一个聚类图(模式空间的若干个精确划分),聚类过程的模糊性主要体现在相似矩阵上,并可以通过模糊信息熵函数度量。聚类过程中模糊性的大小是衡量聚类效果好坏的一个重要指标。降低聚类的模糊性,有利于最终的决策(指定一个精确的划分)。论文引入了交叉熵的概念,通过学习权重,极小化交叉熵,可以有效地降低聚类的模糊性。

According to the transitive closure of a similarity matrix,a dynamic clustering graph which contains several partitions of the sets of objects can be generated.Fuzziness exits in the clustering process and it mainly results from the similarity matrix.The fuzziness of the similarity matrix can be measured by the fuzziness entropy.The less is the fuzzi-ness of the similarity matrix,the more easy is to make the decision for the clustering.In this paper,the authors intro-duce the concept of crossentropy.By minimizing the crossentropy function,the authors can effectively reduce the fuzzi-ness of clustering.