de.mpg.escidoc.pubman.appbase.FacesBean
English
 
Help Guide Disclaimer Contact us Login
  Advanced SearchBrowse

Item

ITEM ACTIONSEXPORT

Released

Report

Overlap-aware global df estimation in distributed information retrieval systems

MPS-Authors
http://pubman.mpdl.mpg.de/cone/persons/resource/persons44113

Bender,  Matthias
Databases and Information Systems, MPI for Informatics, Max Planck Society;

http://pubman.mpdl.mpg.de/cone/persons/resource/persons45041

Michel,  Sebastian
Databases and Information Systems, MPI for Informatics, Max Planck Society;

http://pubman.mpdl.mpg.de/cone/persons/resource/persons45720

Weikum,  Gerhard
Databases and Information Systems, MPI for Informatics, Max Planck Society;

Locator
There are no locators available
Fulltext (public)

MPI-I-2006-5-001.pdf
(Any fulltext), 571KB

Supplementary Material (public)
There is no public supplementary material available
Citation

Bender, M., Michel, S., Weikum, G., & Triantafilou, P.(2006). Overlap-aware global df estimation in distributed information retrieval systems (MPI-I-2006-5-001). Saarbrücken: Max-Planck-Institut für Informatik.


Cite as: http://hdl.handle.net/11858/00-001M-0000-0014-6719-8
Abstract
Peer-to-Peer (P2P) search engines and other forms of distributed information retrieval (IR) are gaining momentum. Unlike in centralized IR, it is difficult and expensive to compute statistical measures about the entire document collection as it is widely distributed across many computers in a highly dynamic network. On the other hand, such network-wide statistics, most notably, global document frequencies of the individual terms, would be highly beneficial for ranking global search results that are compiled from different peers. This paper develops an efficient and scalable method for estimating global document frequencies in a large-scale, highly dynamic P2P network with autonomous peers. The main difficulty that is addressed in this paper is that the local collections of different peers may arbitrarily overlap, as many peers may choose to gather popular documents that fall into their specific interest profile. Our method is based on hash sketches as an underlying technique for compact data synopses, and exploits specific properties of hash sketches for duplicate elimination in the counting process. We report on experiments with real Web data that demonstrate the accuracy of our estimation method and also the benefit for better search result ranking.