模糊加权多视角可能性聚类算法  

MULTI-VIEW POSSIBILITY CLUSTERING ALGORITHM USING FUZZY WEIGHTING

作　　者：王振辉[1] 夏鸿斌[1] 

机构地区：[1]江南大学数字媒体学院,江苏无锡214122

出　　处：《计算机应用与软件》2017年第4期294-298,328,共6页Computer Applications and Software

基　　金：江苏省自然科学基金重点研究专项项目(BK2011003)

摘　　要：受益于独有的可能性聚类特性,较之传统FCM、k-means等基于类均值方法,PCM拥有更佳的聚类效果和抗噪性能。但PCM为传统单视角聚类算法,其在面对新兴多视角聚类场景时,往往效果欠佳。为解决此问题,基于PCM,提出一种新型的称为模糊加权多视角可能性聚类WCo-PCM算法。WCo-PCM显著优点在于其具备对各视角的自适应加权。有关UCI数据集的实验结果表明该算法较传统聚类算法及多视角聚类算法更具抗干扰性,有着更佳的聚类性能。Benefiting from the delicate mechanism of possibility clustering, PCM appears preferable performance regarding effectiveness and anti-noise, against those conventional mean-based methods such as FCM and k-mean. However, PCM still belongs to the traditional single-view clustering method, which incurs its inefficiency in the fashionable multi-view-oriented scenario. For addressing such challenge, based on PCM, a novel clustering algorithm, referred to as fuzzily weighted multi-view Co-PCM （WCo-PCM for short）, is proposed in this paper. The distinctive merit of WCo-PCM lies in its self-adaptive weighting mechanism for the multiple views. The experimental studies implemented on some UCI data sets indicate that, compared with some traditional clustering approaches as well as some existing multi-view ones, WCo-PCM features better anti-interference and clustering effectiveness.

关 键 词：可能性聚类 多视角聚类 模糊加权 抗干扰性 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]