License: Creative Commons Attribution 3.0 Unported license (CC BY 3.0)
When quoting this document, please refer to the following
DOI: 10.4230/LIPIcs.SEA.2017.15
URN: urn:nbn:de:0030-drops-76183
Go to the corresponding LIPIcs Volume Portal

Baumstark, Niklas ; Gog, Simon ; Heuer, Tobias ; Labeit, Julian

The Quantile Index - Succinct Self-Index for Top-k Document Retrieval

LIPIcs-SEA-2017-15.pdf (0.6 MB)


One of the central problems in information retrieval is that of finding the k documents in a large text collection that best match a query given by a user. A recent result of Navarro & Nekrich (SODA 2012) showed that single term and phrase queries of length m can be solved in optimal O(m+k) time using a linear word sized index. While a verbatim implementation of the index would be at least an order of magnitude larger than the original collection, various authors incrementally improved the index to a point where the space requirement is currently within a factor of 1.5 to 2.0 of the text size for standard collections.

In this paper, we propose a new time/space trade-off for different top-k indexes. This is achieved by sampling only a quantile of the postings in the original inverted file or suffix array-based index. For those queries that cannot be answered using the sampled version of the index we show how to compute the query results on the fly efficiently. As an example, we apply our method to the top-k framework by Navarro & Nekrich. Under probabilistic assumptions that hold for most standard texts, and for a standard scoring function called term frequency, our index can be represented with only sublinearly many bits plus the space needed for a compressed suffix array of the text, while maintaining poly-logarithmic query times. We evaluate our solution on real-world datasets and compare its practical space usage and performance against state-of-the-art implementations. Our experiments show that our index compresses below the size of the original text. To our knowledge it is the first suffix array-based text index that is able to break this bound in practice even for non-repetitive collections, while still maintaining reasonable query times of under half a millisecond on average for top-10 queries.

BibTeX - Entry

  author =	{Niklas Baumstark and Simon Gog and Tobias Heuer and Julian Labeit},
  title =	{{The Quantile Index - Succinct Self-Index for Top-k Document Retrieval}},
  booktitle =	{16th International Symposium on Experimental Algorithms (SEA 2017)},
  pages =	{15:1--15:14},
  series =	{Leibniz International Proceedings in Informatics (LIPIcs)},
  ISBN =	{978-3-95977-036-1},
  ISSN =	{1868-8969},
  year =	{2017},
  volume =	{75},
  editor =	{Costas S. Iliopoulos and Solon P. Pissis and Simon J. Puglisi and Rajeev Raman},
  publisher =	{Schloss Dagstuhl--Leibniz-Zentrum fuer Informatik},
  address =	{Dagstuhl, Germany},
  URL =		{},
  URN =		{urn:nbn:de:0030-drops-76183},
  doi =		{10.4230/LIPIcs.SEA.2017.15},
  annote =	{Keywords: Text Indexing, Succinct Data Structures, Top-k Document Retrieval}

Keywords: Text Indexing, Succinct Data Structures, Top-k Document Retrieval
Collection: 16th International Symposium on Experimental Algorithms (SEA 2017)
Issue Date: 2017
Date of publication: 07.08.2017

DROPS-Home | Fulltext Search | Imprint | Privacy Published by LZI