Lehrgebiet Informationssysteme

FB Informatik

LG IS

AG DBIS

(C) AG DBIS

Demonstration of Index Techniques for Similarity-based Search in ORDBMSs

Michael P. Haustein

Kaiserslautern University of Technology
Dept. of Computer Science (AG DBIS)
P.O. Box 3049, 67653 Kaiserslautern, Germany
e-mail: haustein@informatik.uni-kl.de

Wolfgang Mahnke

Kaiserslautern University of Technology
Dept. of Computer Science (AG DBIS)
P.O. Box 3049, 67653 Kaiserslautern, Germany
e-mail: mahnke@informatik.uni-kl.de

Norbert Ritter

Arbeitsbereich VSIS
Fachbereich Informatik
Universität Hamburg
Vogt-Kölln-Strasse 30
D-22527 Hamburg
ritter@informatik.uni-hamburg.de

Full paper (PDF version)

Abstract

Today similarity-based search is used in numerous fields of applications like e-commerce, case-based reasoning, knowledge management, or text and image retrieval. To realize a similarity-based search in ORDBMSs, concepts and mechanisms are needed calculating the similarity of a comparison instance and the stored objects. Due to extremely high cost of function calls during query processing and the need to fetch all objects of the search space for calculating similarity values in order to rank the query results, it is essential to offer an index access for similarity-based queries to reduce response times.
In this demonstration, we present local indices for symbolic, numeric, and string attributes, and show the calculation of similarity values for entire objects. These indices support a new way for direct calculation of similarity values also considering table structures without having to access the actual data objects. This can lead to enormous performance benefits.

Technical Report of the 20th British National Conference on Databases (BNCOD 2003), Pages 1-3, Coventry, U.K., July 2003