基于引力相似度和相对密度的不确定数据流聚类

Uncertain Data Stream Clustering Algorithm Based on Gravity Similarity and Relative Density Techniques

针对不确定数据流聚类问题,提出一种基于引力相似度和相对密度的聚类算法.采用在线/离线两阶段处理框架,综合考虑元组之间的相似度与元组自身的不确定性,利用引力相似度为每个不断到达的数据元组寻找可能归属的微簇,以新的离群点处理和在线维护机制来适应数据流的演化情况,并在离线层使用相对密度算法进行聚类,不需要预先指定聚类数且可处理任意形状的微簇.实验结果表明,与现有的聚类方法相比,所提出的算法具有更高的聚类质量和准确度.

For the issue of uncertain data stream clustering,an effective clustering algorithm based on gravity similarity and relative density technique was proposed in this paper.The algorithm adopted an online/offline two-stage processing framework and considered simularity and data uncertainty together to measure the clustering quality.For each incoming tuples,it used gravity similarity to find the possible micro-cluster.Besides,a novel outlier processing and online maintenance mechanism were developed to adapt to the evolution of the data stream.At the offline stage,it used a relative density clustering algorithm to handle arbitrary shape micro clusters.The experimental results show that the proposed algorithm outperforms existing methods in quality and accuracy.