摘要: 近邻传播聚类在计算过程中需构建相似度矩阵,该矩阵的规模随样本数急剧增长,限制了算法在大规模数据集上的直接应用。为此,提出一种改进的近邻传播聚类算法,利用数据点的局部分布,借鉴半监督聚类的思想构造稀疏化的相似度矩阵,并对聚类结果中的簇代表点再次或多次聚类,直至得到合适的簇划分。实验结果表明,该算法在处理能力和运算速度上优于原算法。
关键词:
近邻传播聚类,
大规模数据集,
数据挖掘
Abstract: Affinity Propagation(AP)clustering takes the full similarity matrix to perform propagation, which limits its application in large scale dataset. An improved affinity propagation clustering is proposed specially for processing large dataset, which fully utilizes local distribution to add constraint like semisupervised clustering to construct sparse similarity matrix. AP runs on sparse similarity matrix to obtain an initial cluster partition, and runs iteratively on the exemplars until it obtains a reasonable partition. Experimental results demonstrate that improved affinity propagation performs better both in processing scale and processing time.
Key words:
affinity propagation clustering,
large scale dataset,
data mining
中图分类号:
谷瑞军, 汪加才, 陈耿, 陈圣磊. 面向大规模数据集的近邻传播聚类[J]. 计算机工程, 2010, 36(23): 22-24.
GU Rui-Jun, HONG Jia-Cai, CHEN Geng, CHEN Ku-Lei. Affinity Propagation Clustering for Large Scale Dataset[J]. Computer Engineering, 2010, 36(23): 22-24.