面向海量数据流的基于密度的簇结构挖掘算法

提出一种基于密度的簇结构挖掘算法(mining density-based clustering structure over data streams,简称MCluStream),以解决数据流密度聚类中输入参数选择困难和重叠簇识别等问题.首先,设计了一种树拓扑CR-Tree索引结构,将直接核心可达的一对数据点映射成树结构中的父子关系,蕴含了数据点依赖关系的CR-Tree涵盖了一系列subEps参数下的基于密度的簇结构;其次,MCluStream算法采用滑动窗口的方式更新CR-Tree,在线维护当前窗口上的簇结构,实现了对海量数据流的快速演化聚类分析;再次,设计了一种快速从CR-Tree提取簇结构的方法,根据可视化的簇结构,选择合理的聚类结果;最后,在真实和合成海量数据上的实验验证了MCluStream算法具有有效的挖掘效果、较高的聚类效率和较小的空间开销.MCluStream可适用于海量数据流应用中自适应的密度聚类演化 分析.

Density-Based Cluster Structure Mining Algorithm for High-Volume Data Streams

This paper proposes a mining algorithm of density-based cluster-structure, named MCluStream, to resolve the problems of input parameter selection and overlapping cluster identification in evolving data stream. First, a tree topology index, named CR-Tree, is designed to map a pair of data points with directly core reachable into relationship of father and child node. The CR-Tree that record relationships among points represents cluster-structure under a series of subEps settings. Second, the online update of cluster-structure on CR-Tree is completed by MCluStream under sliding window environments, which effectively maintains clusters over massive evolving data streams. Third, a fast cluster-structure extraction method is implemented from the CR-Tree. Users can easily select reasonable clustering results according to the visualized cluster-structure. Finally, experimental evaluations on massive-scale real and synthetic data demonstrate the effective mining result and better performance of the proposed algorithm compared against state-of-the-art methods. MCluStream is desirable to be applied to self-adaptive density-based clustering over high-volume data streams.