Exploiting correlation to rank database query results

Jaehui Park, Sang Goo Lee

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Scopus citations

Abstract

In recent years, effective ranking strategies for relational databases have been extensively studied. Existing approaches have adopted empirical term-weighting strategies called tf×idf (term frequency times inverse document frequency) schemes from the field of information retrieval (IR) without careful consideration of relational model. This paper proposes a novel ranking scheme that exploits the statistical correlations, which represent the underlying semantics of the relational model. We extend Bayesian network models to provide dependence structure in relational databases. Furthermore, a limited assumption of value independence is defined to relax the unrealistic execution cost of the probabilistic model. Experimental results show that our model is competitive in terms of efficiency without losing the quality of query results.

Original languageEnglish
Title of host publicationDatabase Systems for Advanced Applications - 16th International Conference, DASFAA 2011, Proceedings
Pages296-305
Number of pages10
EditionPART 2
DOIs
StatePublished - 2011
Event16th International Conference on Database Systems for Advanced Applications, DASFAA 2011 - Hong Kong, China
Duration: 22 Apr 201125 Apr 2011

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
NumberPART 2
Volume6588 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference16th International Conference on Database Systems for Advanced Applications, DASFAA 2011
Country/TerritoryChina
CityHong Kong
Period22/04/1125/04/11

Keywords

  • Attribute value
  • Bayesian networks
  • Correlation
  • Keyword search over structured data
  • Ranking

Fingerprint

Dive into the research topics of 'Exploiting correlation to rank database query results'. Together they form a unique fingerprint.

Cite this