200 likes | 328 Views
Scalable Skyline Computation Using Object-based Space Partitioning. Shiming Zhang Nikos Mamoulis David W. Cheung sigmod 2009. Outline. Introduction Object-based Space Partitioning Recursive Object-based Space Partitioning Left-Child/Right-Sibling Skyline Tree OSPSOnSortingFirst
E N D
Scalable Skyline Computation Using Object-based Space Partitioning ShimingZhang Nikos Mamoulis David W. Cheung sigmod 2009
Outline • Introduction • Object-based Space Partitioning • Recursive Object-based Space Partitioning • Left-Child/Right-Sibling Skyline Tree • OSPSOnSortingFirst • OSPSOnPartitioningFirst • FilterDominatedPartitions • Experimental Results • Conclusions
Introduction(1) • Skyline queries are useful in multi-criteria decision making applications that involve high dimensional and large datasets. • There is a number of methods that operate on pre-computed indexes on the data. • Compare each accessed point with the skyline points found so far.
Introduction(2) • Object-based space partitioning(OSP) scheme, which recursively divides the d-dimensional space into separate partitions w.r.t. a reference skyline object. • Organizes the current skyline points in a search tree. • Object o dominates another object o' iff o is not worse than o' in all dimensions and better than o' in at least one dimension.
Object-based Space Partitioning reference skyline
Recursive Object-based Space Partitioning reference skyline
Why can safely skip? • Skip all incomparable partitions according to Corollary 1
Experimental Evaluation • Three types of synthetic datasets • anti-correlated (AC) • NBA • uniform and independent (UI) • Household • correlated (CO) • Color
Conclusions • Proposed an efficient set of skyline evaluation algorithms that are based on the idea of organizing the discovered skyline points in a tree. • Each candidate skyline object only needs to be compared for dominance with a small subset of the existing skyline points. (skip incomparable sets ) • Makes our solutions scalable to the dimensionality, a feature that all previously proposed skyline algorithms lack.