4The Role of Semantic Relevance in Dynamic User Co

A.BanksPidducketal.(Eds.):CAISE2002,LNCS2348,pp.200-215,2002.Springer-VerlagBerlinHeidelberg2002TheRoleofSemanticRelevanceinDynamicUserCommunityManagementandtheFormulationofRecommendations*NickPapadopoulos1andDimitrisPlexousakis1,21InstituteofComputerScience,FoundationforResearchandTechnology-HellasP.O.Box1385,GR-71110,Heraklion,Greece{npap,dp}@ics.forth.gr2DepartmentofComputerScience,UniversityofCreteP.O.Box2208,GR-71409,Heraklion,GreeceAbstract.Inrecentyears,anincreasinginterestinrecommendationsystemshasemergedbothfromtheresearchandtheapplicationpointofviewandinbothacademicandcommercialdomains.Themajorityofcomparisontechniquesusedforformulatingrecommendationsarebasedonset-operationsoveruser-suppliedtermsorinternalproductcomputa-tionsonvectorsencodinguserpreferences.Inbothcaseshowever,theidentical-nessoftermsisexaminedratherthantheiractualsemanticrelevance.Thispaperproposesarecommendationalgorithmthatisbasedonthemaintenanceofuserprofilesandtheirdynamicadjustmentaccordingtotheusers’behavior.Moreover,thisalgorithmreliesonthedynamicmanagementofcommunities,whichcontainsimilarandrelevantusersandwhicharecreatedaccordingtoaclassificational-gorithm.Thealgorithmisimplementedontopofacommunityman-agementmechanism.Thecomparisonmechanismusedinthecontextofthisworkisbasedonsemanticrelevancebetweenterms,whichisevalu-atedwiththeuseofaglossaryofterms.1IntroductionInrecentyears,anincreasinginterestinrecommendationsystemshasemergedbothfromtheresearchandtheapplicationpointsofview,andinbothacademicandcom-mercialdomains.Manyonline“e-shops”haveadoptedrecommendationtechniquestorecommendnewitemstotheircustomersbasedonaloggedhistoryofpreviouspur-chasesortransactions.Themajorityofexistingrecommendationsystemsdoesnotadequatelyaddresstheinformationfilteringneedsoftheirusers.Aprincipalrequire-mentmainreasonforsuchsystemsistheemploymentofamechanismforfiltering*ThisresearchhasbeensupportedbyprojectCYCLADES(IST-2000-25456):AnOpenCol-laborativeVirtualArchiveEnvironment.TheRoleofSemanticRelevanceinDynamicUserCommunityManagement201informationthatisavailablethroughtheWeb,anddynamicallyadjustittotheuser’sneedsandinterests[1].Informationfilteringanditssubsequenttailoringtotheuser’sinterestsconstitutetheultimategoalsofrecommendationsystems.Modernrecom-mendationsystemsmainlybasetheirfunctionalityononeoftwoapproaches,inordertorecommendadocumentorproducttoauser.Thesetwoapproachesarecontent-basedfilteringandcollaborative-filtering[2,3,6,7]respectively.Theadoptionoftheformerorthelatterdependsonthetypeofinformationthatasystemaimstoprovidetoitsusers.Inthecontent-basedapproach,thesystemfiltersinformation(typicallyadocu-ment)andaimstoproviderecommendationsbasedonthecontentsofthisdocument.Hence,giventheirpreferences,recommendationsfordocumentsthatarerelevanttotheirinterestsareforwardedtotheusers[2,3].Suchanapproachhasbeenmainlyadoptedbyinformationretrieval[7,8]andmachinelearningsystems[9,10,11].Forinstance,inthecaseoftextdocuments,recommendationsareprovidedbasedonthedegreeofmatchingbetweenthecontentofadocumentandauser’sprofile.Theuser’sprofileisbuiltandmaintainedaccordingtoananalysisthatisappliedtothecontentsofthedocumentsthattheuserhaspreviouslyrated[11,12].Theuser’sratingsincom-binationwithamechanismforobtainingthefeaturetermsofadocument[2,3,4,5]aretheonlyrequirementofthesesystems.Ontheotherhand,systemsadoptingthecollaborative-filteringapproach,aimtoidentifyusersthathaverelevantinterestsandpreferenceswithaparticularuser.Thereafter,thedocumentsthattheseuserspreferarerecommendedtothatparticularuser.Theideabehindthisapproachisthat,itmaybeofbenefittoone’ssearchforinformationtoconsultthebehaviorofotheruserswhosharethesameorrelevantinterestsandwhoseopinioncanbetrusted[4].Suchsystemstakeadvantageofthedocumentsthatotherusershavealready“discovered”andrated.Inorderforthesesystemstobeableto“detect”therelevancebetweenusers,theremustbeacompari-sonmechanism,whichinthiscasetoo,istheusageofuserprofileinformation[11,12].Therequirementimposedbysuchsystemsisthatnewusershavetoratesomedocuments,sothatprofilescanbebuiltforthem[2,3,4,5].Ifthesetwoapproachesareappliedseparately,theypresentcrucialdisadvantagesandsufferfromspecificproblems.Inorderfortheseproblemstobedealtwith,hybridapproachesneedtobedevised.Motivatedbytheseissues,thispaperpresentssuchahybridapproachinwhichthesemantic(andnotsolelylexical)relevanceoftermsindynamicallymaintaineduserprofilesisexploitedfortheformulationofrecommenda-tions.Therestofthepaperisorganizedasfollows.Section2presentsacomparisonbe-tweencontent-basedfilteringandcollaborativefilteringaswellasthebenefitsofhybridapproaches.Section3describestheproposedalgorithmforformulatingrec-ommendations.Section4presentspreliminaryexperimentalresults.Section5summa-rizesourcontributionanddrawsdirectionsforfurtherresearch.202NickPapadopoulosandDimitrisPlexousakis2ComparingContent-BasedandCollaborativeFilteringArepresentationofthedataavailableinarecommendationsystemisdepictedinFig-ure1.Dataarerepresentedbymeansofamatrixwhoserowscorrespondtousersandcolumnscorrespond