Reconsidering relevance and embracing interaction
Daniel Tunkelang · Bulletin of the American Society for Information Science and Technology · 2009
Weliveinanageofubiquitoussearch.Childrengrowinguptoday inthedevelopedworldwouldprobablyfindaworldwithout searchenginesasinconceivableastherestofuswouldfinda worldwithoutelectricityandrunningwater.Moreover,searchengineusers ‐thatistosay,almostallofus‐expectthoseenginestoworklikemagic, extrapolatingourinformationneedsfromapairofwords(onaverage)and respondingwiththecontentmostrelevanttothoseneeds. Ofcourse,searchenginesdonotalwaysmeettheseexpectations. Thousandsofengineersworkat‐orwith‐searchenginecompaniesto tunethewaysthoseenginesmodelandcomputerelevance.Relevance, however,isnotaone-size-fits-allmeasure.Forexample,websearch enginestypicallymeasuretherelevanceofwebpagesusingbibliometric considerationssuchasGoogle’sPageRankmeasureandthe“anchortext” associatedwithlinksconnectingintopages.Incontrast,enterprisescannot typicallyrelyonarichnetworkconnectingthedocumentsstoredintheir repositories;hence,theirsearchengineslookforrelevancesignalsin documentstructure(forexample,titleandkeywordfields)andconsiderfactors specifictotheirdomainsorapplications.Ofcourse,almosteverysearch engine’sfilteringandrankingalgorithmstakeintoaccounttherelationship betweenwordsinthesearchqueryandwordsthatoccurindocuments.