2010 SIGIR Best Student Paper: A comparison of general vs. personalized affective models for the prediction of topical relevance 可以作为信息检索顶级会议投稿的范文
AComparisonofGeneralvsPersonalisedAffectiveModelsforthePredictionofTopicalRelevance
IoannisArapakis,KonstantinosAthanasakos,JoemonM.Jose
DepartmentofComputingScience
UniversityofGlasgowGlasgow,G128QQ
{arapakis,athanask,jj}@dcs.gla.ac.uk
ABSTRACT
Informationretrievalsystemsfaceanumberofchallenges,originatingmainlyfromthesemanticgapproblem.Implicitfeedbacktechniqueshavebeenemployedinthepasttoad-dressmanyoftheseissues.Althoughthiswasastepto-wardstherightdirection,aneedtopersonaliseandtailorthesearchexperiencetotheuser-speci cneedshasbecomeevident.Inthisstudyweexaminewaysofpersonalisingaf-fectivemodelstrainedonfacialexpressiondata.Usingper-sonaliseddataweadaptthesemodelstoindividualusersandcomparetheirperformancetoageneralmodel.Themaingoalistodeterminewhetherthebehaviouraldi erencesofusershaveanimpactonthemodels’abilitytodeterminetopicalrelevanceandif,bypersonalisingthem,wecanim-provetheiraccuracy.FormodellingrelevanceweextractasetoffeaturesfromthefacialexpressiondataandclassifythemusingSupportVectorMachines.Ourinitialevaluationindicatesthataccountingforindividualdi erencesandap-plyingpersonalisationintroduces,inmostcases,anoticeableimprovementinthemodels’performance.
inacollection.Althoughprogresshasbeenmade,thee ec-tivenessofexistingsystemsisstilllimited.Thegapisfurtherwidenedwhentheuserisdrivenbyanill-de nedinforma-tionneed,oftentheresultofananomalyinhis/hercurrentstateofknowledge[7].Theformulatedsearchqueries,whichareusedbytheretrievalsystemstolocatepotentiallyrel-evantitems,produceresultsthatdonotaddresstheusers’trueneeds.
TodealwithinformationneeduncertaintyIRsystemshaveemployedinthepastarangeoffeedbacktechniques,whichvaryfromexplicit[13,21]toimplicit[1,6].Theno-tionofexplicitfeedbackwaspresentfromtheearlyyearsofIR,butitsoonbecameapparentthatuserscouldnotcopewiththecognitiveburdenofexplicitrelevancejudgments.Alternativepathshadtobediscovered,whichledtotheun-obtrusive,yetlessrobust,implicitfeedbacktechniques[12,15].Eventhoughthiswasasteptowardstherightdirectionaneedtopersonaliseandtailorthesearchexperiencetotheuser-speci cneedswasprogressivelymadeevident.
Personalisationemergedasanappealingtechniqueindeal-ingwiththeissuescausedbythevariationofonlinebe-haviourandtheindividualdi erencesobservedinuserin-terests,informationneeds,searchgoals,di cultiesencoun-tered,andother.ToapplypersonalisationanIRsystemmustinitiallyemployamodellingtechniquethatwillcap-turecertainusercharacteristics.Atalaterstage,infor-mation lteringisperformedtore netheaggregatedinfor-mationandadjustthesystem’sresponsestoaccommodateusers’needs,thusprovidingamorepersonalisedexperience.Severalattemptshavebeenmadeinthepasttodevelopusermodels,usingimplicitfeedback.In[16],OardandKimde neasetofapplication-speci cobservablebehaviours(ex-amination,retention,etc.)andintroducetheconceptoflearninguserinterestsandbuildinguserpro lesfromim-plicitdata.In[20],Puolam¨akietal.combineimplicitfeed-backwithexplicitlycreateduserpro les.Inthelatterwork,theauthorsusemixturemodelstocombinedi erentsourcesofrelevancejudgments.Theimplicitfeedbackinformationderivesfromeye-movementdata,usedincombinationwithaprobabilisticcollaborative lteringmodel.
In[2]theauthorsmaketheassumptionthat,apartfromusermodelling,query-speci cbehavioursarealsoimportantandshouldbeconsideredwhenattemptingtopredicttopicalrelevance.Followingthiswork,Liuetal.[14]constructeduserpro lesbasedonusers’searchhistoryanddevelopedal-gorithmsthatmappedquerytermstoprede nedcategories.Thelatterinformationwasusedtoextractusers’interestsandaddressissuesrelatedtowordambiguity.Teevanet
CategoriesandSubjectDescriptors
H.3.3[InformationSearchandRetrieval]:RelevanceFeedback,SearchProcess;I.5.1[ComputingMethodolo-gies]:PatternRecognition—Models
GeneralTerms
Experimentation,HumanFactors,Performance
1.INTRODUCTION
Themainchallengeinformationretrieval(IR)systemsfacenowadaysoriginatesfromthesemanticgapproblem:thesemanticdi erencebetweenauser’squeryrepresenta-tionandtheinternalrepresentationofaninformationitemTheresearchleadingtothispaperwassupportedbytheEuropeancommission,underthecontractFP6-027122(SALERO).
Permissiontomakedigitalorhardcopiesofallorpartofthisworkforpersonalorclassroomuseisgrantedwithoutfeeprovidedthatcopiesarenotmadeordistributedforpro torcommercialadvantageandthatcopiesbearthisnoticeandthefullcitationonthe rstpage.Tocopyotherwise,torepublish,topostonserversortoredistributetolists,requirespriorspeci cpermissionand/orafee.
SIGIR’10,July19–23,2010,Geneva,Switzerland.
Copyright2010ACM978-1-60558-896-4/10/07...$10.00.
2010 SIGIR Best Student ... 8页 免费 WWW 2007 Poster Paper T... 2页 免费 2009 SIGIR Best Paper So... 8页 免费 2010 EST best policy pap......
2010 SIGIR Best Student ... 8页 免费 2010 SIGIR Best Paper As... 8页 免费 2010.03季度农 暂无评价 4页 免费 2010试卷(03) 暂无评价 1页 1财富值...
2010 SIGIR Best Paper As... 8页 免费 2010 SIGIR Best Student ... 8... sigir 2010sigir 2010隐藏>> Estimating Advertisability of Tail Queries for ...
The Conclusion for SIGIR 2011_工学_高等教育_教育专区。SIGIR 2011... ZJU 身体的重要性 Tight Schedule From day to night One paper after ...
dl_10_best_certs(2010年... 3页 1下载券 2010 SIGIR Best Paper ... ...Best of Ixiqi - 2010 年度最佳之 10 佳产品 基本上每一年,爱稀奇在年终的...
2010 SIGIR Best Paper As... 8页 免费 Best Hospitals Methodolo... 暂无评价 44页 1财富值 Best US Hospitals list 暂无评价 5页 1财富值 Cataract_in_...
Best of Ixiqi - 2010年度... 8页 免费 Best Practices for HAZOP... 9页 2财富值 2010 SIGIR Best Student ... 8页 免费喜欢此文档的还喜欢...
※ The ”Best Paper “Award (SIGIR) 2011 ※ Area Chair --- SIGIR Qiaoling Liu PhD Student (Curriculum Vitae) Mathematics and Computer Science Department...
He received the Best Paper and Best Demonstration...TAO MEI’S CV *: The student I mentored when...” ACM SIGIR Conference on Research and ...
He received the Best Paper and Best Demonstration...TAO MEI’S CV *: The student I mentored when...” ACM SIGIR Conference on Research and ...
热门文档