Efficient semantic search on DHT overlays

  • Authors:
  • Yingwu Zhu;Yiming Hu

  • Affiliations:
  • Department of CSSE, Seattle University, Seattle, WA 98122, USA;Department of ECECS, University of Cincinnati, Cincinnati, OH 45221, USA

  • Venue:
  • Journal of Parallel and Distributed Computing
  • Year:
  • 2007

Quantified Score

Hi-index 0.00

Visualization

Abstract

Distributed hash tables (DHTs) excel at exact-match lookups, but they do not directly support complex queries such as semantic search that is based on content. In this paper, we propose a novel approach to efficient semantic search on DHT overlays. The basic idea is to place indexes of semantically close files into same peer nodes with high probability by exploiting information retrieval algorithms and locality sensitive hashing. A query for retrieving semantically close files is answered with high recall by consulting only a small number (e.g., 10-20) of nodes that stores the indexes of the files semantically close to the query. Our approach adds only index information to peer nodes, imposing only a small storage overhead. Via detailed simulations, we show that our approach achieves high recall for queries at very low cost, i.e., the number of nodes visited for a query is about 10-20, independent of the overlay size.