Discernability and Preference in Interactive Option Searches
2009 (English)In: The Swedish AI Society Workshop / [ed] Fredrik Heintz, Jonas Kvarnström, 2009, 21-26 p.Conference paper (Refereed)
In option searches, a user seeks to locate an idealoption (e.g. a flight, restaurant, book, etc.) from aset of n such options. The aim of this paper is toprovide a solid mathematical basis for optimizingpresentation length in such searches. The paperdevelops an information theoretic model that takesinto account the user’s ability to discern among optionsas well as their a priori preference. The developedmodel makes definite predictions about whatclusterings of a user query are more or less informativebased on measures of information gain. Usersare offered descriptions of such clusters as the basisfor subsequent refinement steps in a drill-downdialogue to locate the best option. We have implementedan initial system that performs reasonablywell on moderately large data sets and gives intuitivelyappealing results. The system is in theprocess of being integrated into a natural languageinterface system for end-user evaluation.
Place, publisher, year, edition, pages
2009. 21-26 p.
Computer and Information Science
IdentifiersURN: urn:nbn:se:kth:diva-66428OAI: oai:DiVA.org:kth-66428DiVA: diva2:484001
Proceedings of Swedish Artificial Intelligence Society (SAIS). Linköping, Sweden. May 27-28, 2009
QC 201204262012-01-262012-01-262016-06-20Bibliographically approved