基于邻域相似度的近邻传播聚类算法  

Affinity propagation clustering algorithm based on neighborhood similarity

作　　者：赵昱 陈琴[1] 苏一丹[1] 陈慧姣 ZHAO Yu, CHEN Qin , SU Yi-dan, CHEN Hui-jiao(School of Computer and Electronics Information, Guangxi University, Nanning 530004, Chin)

机构地区：[1]广西大学计算机与电子信息学院,广西南宁530004

出　　处：《计算机工程与设计》2018年第7期1883-1888,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61363027)

摘　　要：针对传统近邻传播聚类算法(affinity propagation clustering algorithm,AP)处理特征复杂数据时聚类准确率较低的问题,提出一种基于邻域相似度的近邻传播聚类算法。通过分析数据样本统计特性,确定合适的邻域半径和邻域密度,计算邻域相似度并注入偏向参数,提高算法在特征复杂数据集上的聚类精度。在UCI数据集上的实验结果表明,所提算法的聚类精度优于相比较的AP算法,且邻域半径对不同数据集有自适应性,引入邻域相似度提高传统AP算法在特征复杂数据集上的聚类精度是可行的。Aiming at the problems of the classic affinity propagation（AP）clustering algorithm,such as the poor clustering effect on complex structure data sets,an AP algorithm based on the neighborhood similarity was proposed.By analyzing the statistical characteristics of data samples,the neighborhood radius and neighborhood density were determined,the neighborhood similarity was calculated,and was injected into the AP＇s preference,which improved the clustering accuracy of the algorithm on the irregular shape and complex data.Experimental results on the UCI data sets show that the clustering accuracy of the proposed algorithm is better than the classic AP,and its neighborhood radius is adaptive to different data sets.It is feasible using the neighborhood similarity to improve the classic AP algorithm.

关 键 词：近邻传播聚类算法 偏向参数 邻域半径 邻域密度 邻域相似度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]