Focus and element length for book and wikipedia retrieval

  • Authors:

  • Jaap Kamps;Marijn Koolen

  • Affiliations:

  • Archives and Information Studies, Faculty of Humanities and ISLA, Faculty of Science, University of Amsterdam;Archives and Information Studies, Faculty of Humanities, University of Amsterdam

  • Venue:
  • INEX'10 Proceedings of the 9th international conference on Initiative for the evaluation of XML retrieval: comparative evaluation of focused retrieval
  • Year:
  • 2010

Quantified Score

Hi-index 0.00

Visualization

Abstract

In this paper we describe our participation in INEX 2010 in the Ad Hoc Track and the Book Track. In the Ad Hoc track we investigate the impact of propagated anchor-text on article level precision and the impact of an element length prior on the within-document precision and recall. Using the article ranking of an document level run for both document and focused retrieval techniques, we find that focused retrieval techniques clearly outperform document retrieval, especially for the Focused and Restricted Relevant in Context Tasks, which limit the amount of text than can be returned per topic and per article respectively. Somewhat surprisingly, an element length prior increases within-document precision even when we restrict the amount of retrieved text to only 1000 characters per topic. The query-independent evidence of the length prior can help locate elements with a large fraction of relevant text. For the Book Track we look at the relative impact of retrieval units based on whole books, individual pages and multiple pages.