Automatic Query Type Identification Based on Click Through Information

《Automatic Query Type Identification Based on Click Through Information》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、AutomaticQueryTypeIdentificationBasedonClickThroughInformationYiqunLiu1,MinZhang1,LiyunRu2,andShaopingMa11StateKeyLabofIntelligentTech.&Sys.,TsinghuaUniversity,Beijing,Chinaliuyiqun03@mails.tsinghua.edu.cn2SogouIncorporation,Beijing,Chinaruliyun@sohu-rd.comAbstract.Wereportonastudythatwasu

2、ndertakentobetteridentifyusers’goalsbehindwebsearchqueriesbyusingclickthroughdata.Basedonuserlogswhichcontainover80millionqueriesandcorrespondingclickthroughdata,wefoundthatquerytypeidentificationbenefitsfromclickthroughdataanalysis;whileanchortextinformationmaynotbesousefulbecauseitisonlya

3、ccessibleforasmallpart(about16%)ofpracticaluserqueries.Wealsoproposedtwonovelfeaturesextractedfromclickthroughdataandadecisiontreebasedclassificationalgorithmforidentifyinguserqueries.Ourexperimentalevaluationshowsthatthisalgorithmcan1correctlyidentifythegoalsforabout80%websearchqueries.1I

4、ntroductionWebSearchengineiscurrentlyoneofthemostimportantinformationaccessandmanagementtoolsforWWWusers.Mostusersinteractwithsearchengineusingshortquerieswhicharecomposedof4wordsorevenfewer.Thisphenomenaof”shortqueries”haspreventedsearchenginesfromfindingusers’informationneedsbehindtheirq

5、ueries.Withanalysisintosearchengineuserbehavior,Broder[1]andRose[2]in-dependentlyfoundthatsearchgoalsbehinduserqueriescanbeinformational,navigationalortransactional(referedtoasresourcetypebyRose).Furtherex-perimentresultsinTREC[3][4]showedthatinformationalandnavigationalsearchresultsbenefi

6、tfromdifferentkindsofevidences.Craswell[5]andKraaij[6]foundthatanchortextandURLformatofferimprovementtocontent-onlymethodforhomepagefindingtask,whichcoversamajorpartofnavigationaltypequeries.Bharat[7]provedthatinformationaltypesearchesmaybeim-provedusinghyperlinkstructureanalysis.Accordingto

7、theseresearches,ifquerytypecanbeidentifiedforagivenuserquery,retrievalalgorithmcanbeadaptedtothisquerytypeandsearchperformancecanbeimprovedcomparedwithageneralpurposealgorithm.Thatiswhyweshouldidentifyusers’searchgoalsbehindtheirsubmittedqueries.1SupportedbytheChines