Efficient skyline computation over low-cardinality domains

  • Authors:

  • Michael Morse;Jignesh M. Patel;H. V. Jagadish

  • Affiliations:

  • University of Michigan, Ann Arbor, Michigan;University of Michigan, Ann Arbor, Michigan;University of Michigan, Ann Arbor, Michigan

  • Venue:
  • VLDB '07 Proceedings of the 33rd international conference on Very large data bases
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Current skyline evaluation techniques follow a common paradigm that eliminates data elements from skyline consideration by finding other elements in the dataset that dominate them. The performance of such techniques is heavily influenced by the underlying data distribution (i.e. whether the dataset attributes are correlated, independent, or anti-correlated). In this paper, we propose the Lattice Skyline Algorithm (LS) that is built around a new paradigm for skyline evaluation on datasets with attributes that are drawn from low-cardinality domains. LS continues to apply even if one attribute has high cardinality. Many skyline applications naturally have such data characteristics, and previous skyline methods have not exploited this property. We show that for typical dimensionalities, the complexity of LS is linear in the number of input tuples. Furthermore, we show that the performance of LS is independent of the input data distribution. Finally, we demonstrate through extensive experimentation on both real and synthetic databsets that LS can results in a significant performance advantage over existing technqiues.