标题:INFLUENCE POWER-BASED CLUSTERING ALGORITHM FOR MEASURE PROPERTIES IN DATA WAREHOUSE
作者:Ji, Min; Jin, Fengxiang; Li, Ting; Zhao, Xiangwei; Ai, Bo
通讯作者:Ji, M
作者机构:[Ji, Min; Jin, Fengxiang; Li, Ting; Zhao, Xiangwei; Ai, Bo] Shandong Univ Sci & Technol, Geomat Coll, Qingdao 266510, Peoples R China.
会议名称:Joint International Conference on Theory, Data Handling and Modelling in GeoSpatial Information Science
会议日期:MAY 26-28, 2010
来源:JOINT INTERNATIONAL CONFERENCE ON THEORY, DATA HANDLING AND MODELLING IN GEOSPATIAL INFORMATION SCIENCE
出版年:2010
卷:38
页码:224-228
关键词:Influence Power; Hierarchical Tree; Neighbor Function Clustering; Data; Mining; Gravitational Clustering; Nature Clustering
摘要:The data warehouse's fact table can be considered as a multi-dimensional vector point dataset. In this dataset, each point's measure property can be transformed as the influence power against its neighbor points. If one point's measure is larger, it would have more influence power to attract its neighbor points, and its neighbors would have a trend to be absorbed by this point. Being inspired by the Gravitational Clustering Approach (GCA), the paper introduces a new method named IPCA (Influence Power-based Clustering Algorithm) for clustering these vector points. The paper first defines several concepts and names the local strongest power points as Self-Strong Points (SSPs). Using these SSPs as the initial clustering centers, IPCA constructs serials of hierarchical trees which are rooted by these SSPs. Because there are only a few SSPs left, by using each SSPs' influence power, the paper adopts the neighbor function clustering method to define the clustering criteria function, and gives the detail clustering procedure of SSPs. IPCA follows the nature clustering procedure at the micro-level, with a single scan, it can achieve the initial clustering. From the experiment result, we can see that IPCA not only identifies different scale clusters efficiently, but it also can get arbitrary shape clusters easily.
收录类别:CPCI-S
资源类型:会议论文
TOP