Do we mean the same?: disambiguation of extracted keyword queries for database search

  • Authors:

  • Elena Demidova;Irina Oelze;Peter Fankhauser

  • Affiliations:

  • L3S Research Center, Hannover, Germany;L3S Research Center, Hannover, Germany;L3S Research Center, Hannover, Germany

  • Venue:
  • Proceedings of the First International Workshop on Keyword Search on Structured Data
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Users often try to accumulate information on a topic of interest from multiple information sources. In this case a user's informational need might be expressed in terms of an available relevant document, e.g. a web-page or an e-mail attachment, rather than a query. Database search engines are mostly adapted to the queries manually created by the users. In case a user's informational need is expressed in terms of a document, we need algorithms that map keyword queries automatically extracted from this document to the database content. In this paper we analyze the impact of selected document and database statistics on the effectiveness of keyword disambiguation for manually created as well as automatically extracted keyword queries. Our evaluation is performed using a set of user queries from the AOL query log and a set of queries automatically extracted from Wikipedia articles both executed against the Internet Movie Database (IMDB). Our experimental results show that (1) knowledge of the document context is crucial in order to extract meaningful keyword queries; (2) statistics which enable effective disambiguation of user queries are not sufficient to achieve the same quality for the automatically extracted requests.