基于模糊C-means的多视角聚类算法  

Multi-view clustering algorithm based on fuzzy C-means

作　　者：杨欣欣[1] 黄少滨[1] 

机构地区：[1]哈尔滨工程大学计算机科学与技术学院,黑龙江哈尔滨150001

出　　处：《中南大学学报（自然科学版）》2015年第6期2128-2133,共6页Journal of Central South University:Science and Technology

基　　金：国家科技支撑计划项目(2012BAH08B02);哈尔滨工程大学中央高校基本科研业务专项资金资助项目(HEUCFZ1212;HEUCF100603)~~

摘　　要：目前多数多视角聚类算法属于"刚性"划分算法,不适用于处理具有聚簇重叠结构的数据集,为此,提出一种基于模糊C-means的多视角聚类算法(简称FCM-MVC),该算法利用隶属度描述对象与类别的关系,能够更真实地描述具有聚簇重叠结构数据集的聚类结果。FCM-MVC算法同时利用多个视角信息,自动计算每个视角的权重。研究结果表明:FCM-MVC算法能够有效处理具有聚簇重叠结构的数据集;与已有的3种经典的多视角聚类算法相比,该算法获得的聚类精度更高。Considering that most exiting multi-view clustering algorithms focusing on hard-partition clustering methods, which are not suitable for analyzing dataset with overlapping clusters, a multi-view clustering algorithm based on fuzzy C-means （FCM-MVC） was developed. The membership degree was used to describe the relation between objects and clusters, so FCM-MVC algorithm could more truely describe clustering results of dataset with overlapping clusters. FCM-MVC algorithm simultaneously incorporated fearture information in multi-view space and automatically computes weight of each view. The results show that FCM-MVC can analyze overlapping clusters effectively and the precision of clustering results of FCM-MVC are superior to the three representative algorithms.

关 键 词：多视角聚类 模糊C-means 数据挖掘 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]