Preference-Driven Personalization for Flexible Digital Item

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

MultimediaSystemsmanuscriptNo.(willbeinsertedbytheeditor)

BenjaminK¨ohncke·Wolf-TiloBalke

Preference-DrivenPersonalizationforFlexibleDigitalItemAdaptation

AbstractThedeliveryofmultimediacontentoftenneedstheadaptationofthecontentinordertosatisfyuserconstraints.WiththeDigitalItemAdaptationpart,theMPEG-21standardalreadyde nesausefulframe-worktohandlethistask.However,inmodernservice-orientedarchitecturesthefunctionalityofadaptationissplitoverseveralservices.Hence,thecentralinstantiationofasuitableservicechainneedstotackleacomplexmulti-objectiveoptimizationproblem.Inthisproblembetweencontentchoiceandpossibleadaptationsthecurrentpref-erencemodelintheMPEG-7/21standardstilllacksex-pressiveness.Inthecourseofthispaperwedemonstratethisshortcomingandhowtheintegrationofmorepower-fulmodelscaneasetheinstantiationproblem.Further-moreweexplainhowtoe cientlyevaluatepreferencetrade-o sbyevaluatingskylinequeriesascurrentlyin-vestigatedinthe eldofinformationsystems.Asarun-ningexampleweusepreference-basedcontentadapta-tioninatypicalmediastreamingapplicationwithWebservicesasbasicmodules.Thecontributionofourframe-workistoenableacentralcoordinatortoinstantiateanexecutableservicecompositionchainbyintegratingallneededWebservicestoadaptthemultimediacontentinthebestpossiblefashioninthesenseofParetooptimal-ity.

KeywordsMultimediaSystems·DigitalItemAdap-tation·Personalization·SkylineQueries·PreferenceModeling

ThisworkwassupportedinpartbytheGermanResearchFoundation(DFG)withintheEmmy-NoetherProgramofExellence.

Wolf-TiloBalke

L3SResearchCenterattheUniversityofHannover,Han-nover,Germany

Tel.:+49-511-762-17712Fax:+49-511-762-17779E-mail:balke@l3s.de

BenjaminK¨ohncke

L3SResearchCenterattheUniversityofHannover,Han-nover,Germany

E-mail:koehncke@l3s.de

1Introduction

Deliveringmultimediacontentoveraplethoraofde-vicesinapersonalizedfashionputsgreatdemandsontheselectionandadaptationofcontent[24].Therearetwobasicapproaches:eitheraserversideadaptation,wherealladaptedversionsoftheappropriatecontentarethenbroadcastedinastream,oraclientsideadap-tationwherepowerfuldecodershavetobeavailable,toperformanon-the- yadaptation.Buildingonservice-orientedadaptationourapproachgoesbeyondtheseba-sicmodelsandprovidesaninpidualservicechainforeachuserwheretheadaptationisdoneon-the- y.Forev-eryinpidualuseraspeci cwork owhastobecreatedrespectingtheinpidualuser’scontentpreferences,gen-eralcontentsemanticsandnetworkconstraints,aswellasterminalcapabilities.

Thisisbecausetoday,insteadofbuildingcomplexmono-lithicsystemsthe exiblecompositionofservicesintosuitablework owsformediahandlingisanimportantgoalforsystemintegrationandreusabilityofcompo-nents(foranoverviewsee[17]).Suchwork owshavetobeplannedandtheirexecutioncloselymonitored,some-timesneedingtheseamlessintegrationofreplacementsforfailedservices.Tofacilitatethis,theinstantiationoftheservicechainneedstosolveamulti-objectiveopti-mizationproblemconsideringpossibleservices(ashavebeendiscovered,cf.[1]),thecontentcurrentlyavailableandtheauthor’s,user’sandclientdevice’sconstraintsontheadaptation.

Afterhavingaddressedthebasicproblemin[11],inthispaperwediscusshowtoadaptcontentinapersonal-izedfashionbyintegratingMPEG-7/21metadatawithacomplexpreferenceframework,namelyamodelonqual-itativepartialorderpreferences(cf.[6],[9])andshowhowtoevaluatethecomplexresultmeshesgivenbytherespectiveproductorderundertheconceptofParetoop-timality.

Thebene tistwofold:ononehandintegratingamorepowerfulmodelenableshigherexpressivenessofuserpref-erences;ontheotherhandcompositionenginesandser-

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

2BenjaminK¨ohncke,Wolf-TiloBalke

vicesalongthework owchaincandirectlyusetheaug-mentedmetadataforchoosingadaptationstrategiesmore

user(intheTerminalCapabilitiesdescriptionscheme).Figure1showsthetypicalinteractionofacontentadap-intelligently.Everyserviceadaptsthemetadataaccord-ingtothetransformationofthecontentitperformedandremovesobsoletepreferenceinformation.Thisalsoenablesamoree cientmatchingofconstraintsforthoseserviceslaterinaservicecompositionchain.

Asarunningexampleapplicationthroughoutthispaperwewillusemediastreamingthat exiblyadaptssomemultimediacontenttotheterminalcapabilitiesoftheuser.Thus,usersgetthebestpossiblequalitywithre-specttotheirrespectiveterminalcapabilities.Ourproto-typicalimplementationusesWebservicesasbasicmod-ulestobuildmultimediaapplications.ThedescriptionofthecomplexdatatypesisprovidedbyanMPEG-7/21descriptionattachedtothemediadata.EachinpidualWebserviceevaluatestheMPEG-7/21descriptionandadaptsthemultimediamaterialtothespecialneedsoftheuserandtheclientdevice.

Tocoordinatethecompositionwecurrentlyuseaded-icatedcentralWebserviceinstancethattakesoverre-sponsibilityforselectingthebestavailablecontentandasuitablework owbymeansofthetechnicalpro leanduserpreferences[1].Thisserviceinstantiationandmon-itoringservice(SIAM)doesalsomonitorthework ow,butitdoesnotindetaildecideinwhatwayeachWebserviceshouldadaptthemultimediacontent.Everyin-pidualWebservice(e.g.,mergingortranscodingser-vices)canuseitsownheuristicsabouthowtoadapttheselectedmultimediacontentbestwithrespecttothespeci edtargetoutputfortheend-user.Inadditiontothepurelytechnicalinformation,theservicesalsouseim-portantdescriptionschemesfromtheMPEG-7/21stan-dardlikee.g.,transcodinghints[23].

Therestofthepaperisstructuredasfollows:Insec-tion2wewillgiveanoverviewabouttheuserspeci cmetadatadescriptionso eredbyMPEG-7/21.Ashortsummaryaboutwhathasalreadybeendoneinthe eldofpersonalizedmediaadaptationisgivenintherelatedworkpartinsection3.Thenweintroduceamoreexpres-sivepreferencemodelinsection4andpresentthebasicarchitectureofourprototypicalsystem.Insection5wedescribehowtoactuallyevaluatethecomplextrade-o meshofaParetopreferencegraphandtheprototypicalimplementationofamediastreamingtestcaseisdis-cussedinsection6.Finally,weclosethepaperwithourconclusionsandsomefutureresearchdirections.

2MetadataDescriptionsinMPEG-7/21

TheMPEG-7/21UsageEnvironmentisapartoftheMPEG-21DigitalItemAdaptation(DIA)architecture(Part7,ISO/IEC21000-7)ando ersseveralopportu-nitiestode neuserspeci cinformation.ItincludestheUserInteractionToolsfromMPEG-7[23]andalsothedescriptionelementsforthedevicecapabilitiesofthe

tationenginewithamediadatabaseandtheenduser,respectivelyhis/herclientdevice.Mediadata

generally

Fig.1GeneralDigitalItemAdaptation

isretrievedbymatchingmetadatadescriptions.Toper-formacontentselectionthepersonalizedadaptationde-cisionenginematchestheresourcedescriptionsfromametadatadirectoryandtheuser-providedinformation.Thisinformationconsistsoftheactualuserquery,moregeneralcontentpreferencesoftheuser,andtechnicalconstraintsthatprovidedataaboutthetechnicaldeviceoftheuser.Afterdeterminingtheoptimaladaptation,thedecisionishandedontotheadaptationengine.Theadaptationenginethenretrievesthemediaresourcefromthemediadatabase,adaptsitaccordinglyanddeliverstheadaptedresourcetotheuser.Basicallyanadaptationworksasfollows:

1.Thedecisionenginegetsarequestwithcontentpref-erencesandtechnicalconstraints

2.Iftherequestedmediaisavailableandallconstraintsaresatis edtherespectivemultimediadataisdi-rectlydelivered

3.Iftherequestedmediaisavailable,butsomecon-straintsarenotyetsatis ed

3.1.Ifthemultimediadatacanbeadaptedtomeetall

constraints,adaptmediaaccordinglyanddeliver3.2.Ifnot,choosethehighestrankingpreference(s)

toadapttootherformats,resolutions,etc.ortochoosenewcontentHowever,thisadaptationschemecannotworkwithmorecomplextrade-o s,andneithercanitdecideintelligentlywhentochoosealternativeadaptationoptionsthat(pos-siblyiftakentogether)wouldmakemoresensethansim-plythehighestratedpreference.LetustakeacloserlookonwhatMPEG-7/21canactuallyexpressinuserpref-erences.

TheUserInteractionTools(UIT)o erseveralpossibil-itiestodescribecontentpreferencesoftheuser.Incon-trasttotheotherdescriptionschemesde nedinMPEG-7theUITareboundtotheuserandnottothemulti-mediamaterial:theUserPreferencesareusedtoex-pressspeci cpreferencesoftheuserintheselectionof

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

Preference-DrivenPersonalizationforFlexibleDigitalItemAdaptation3

multimediamaterialandconsistofseveraldescriptionschemes:

Todecidebetweenpreferencesallthepreferenceval-ueswithineachattributearecomparedseparately:thehigherthevalue,themoreimportanttherespectivepref-–AsetofFilteringandSearchPreferencesdescribeserence.However,theMPEG-7/21standardjustde nesinpidualuserwisheswithrespecttothe lteringthesyntaxandsemanticsoftheuserpreferencedescrip-andsearchingofmultimediadata.Eachelementistionscheme,butnottheextractionmethodofthepref-decomposedinto:erencevalue,cf.[12].Thatmeansallusershavetoas-–FilteringandSearchPreferenceelementsonalowersignpreferencevaluesmanually.Butcansemanticallyleveltoallowforpreferencehierarchies.incomparableattributes(likepreferencesonactorsand–Classi cationPreferenceelementstodescribepref-languagesettings)becomparedinaquantitativeway?erencesregardingtoattributes,suchaslanguage,productionformatorcountryoforigin.–CreationPreferenceelementstodescribethepre-ferredcreationofcontent,suchastitle,creatororcreationdate.–SourcePreferenceelementswherepreferredrepos-itoriestoretrievemultimediamaterialfrom(e.g.,mediaserversordigitallibraries)canbede ned.–PreferenceConditionelementswheretheusercanconstraintheapplicabilityofspeci c lteringandsearchpreferences(forexampletimeanddate).–AsetofBrowsingPreferenceselementsdescribetheuser’swishesregardingmultimediacontentnaviga-tionandbrowsing.AtypicalexampleisaSummaryPreferencedescribingthepreferredtypeofcontentsummaries.Iftheusergotmorethanonepreference,aweightingfac-tor,calledPreferenceValue,canbespeci edtoexpress

therelativeimportanceofapreference.Thepreference

valueisasimplenumericalvaluerangingfrom-100to

100.Thevalueindicatesthedegreeofuserspreferenceor

non-preference.Thezerovalueindicatesthattheuseris

neutralintermsofpreferenceversusdislike.Adefault,

positive,valueof10correspondstoanominalpreference.

Bychoosinganegativeweightingtheusercanexpress

negativepreferencesordislikes.

Example:ThefollowingXMLcodesnippetshowsacomplexpreferenceinMPEG-21notationforuser’Kim’,whoprefers’Action’moviesstarring’MattDamon’overtheactors’ArnoldSchwarzenegger’and’BradPitt’andonthewholemoviesthatareinGermanlanguageovermoviesinEnglishlanguageusingappropriateFilteringandSearchPreferenceelements.<UserPreference>

<UserIdentifieruserName="Kim"/>

<UsagePreference><FilteringAndSearchPreferences>

<ClassificationPreference><Genre>Action</Genre><LanguagepreferenceValue="90">german</Language>

<LanguagepreferenceValue="70">english</Language>

</ClassificationPreference><CreationPreference><ActorpreferenceValue="63">Damon</Actor>

<ActorpreferenceValue="50">Schwarzenegger</Actor>

<ActorpreferenceValue="27">Pitt</Actor></CreationPreference>

</FilteringAndSearchPreference></UsagePreference></UserPreference>

Ithardlymakessensetostatesomethinglike:’IpreferMattDamonmoviestomoviesinGerman’.Moreover,fortheuseritisentirelyunintuitive,whataninpidualpreferencevalue(like63or50)actuallymeans.Otherau-thors,e.g.[27],haveproposedtouseevenmorecomplexutilityfunctions,butafterallstillrelyonquiteunintu-itivequantitativepreferenceframeworks.

Intheexampleabovewehavetwopreferencesondi er-entattributes:languageandactors,buttheattributesarebasicallyincomparable.Thus,somecombinationsformediaobjectsbecomealsoincomparableandcannotberankedinatotalorder.ThischaracteristicleadsdirectlytotheconceptofParetooptimality:theParetoset(oftenalsocallede cientfrontier)consistofallnon-dominatedobjects,i.e.foreachobjectnootherobjectinthesethasbetteroratleastequalattributevalueswithre-specttoallattributes.ConsiderforinstanceanEnglishMattDamonmovieandaGermanArnoldSchwarzeneg-germovie.Bothareincomparable,becauseoneisbet-terwithrespecttothelanguagepreference,whereasthe

otherisbetterwithrespecttotheactorpreference.How-ever,bothoptionsdominateanEnglishBradPittmovie,

whichaccordinglywouldnotbepartoftheParetoset.

TheuseoftheParetosemanticsisalsoadvocatedin[14]

providingadecisiontakingframeworkwherehard-and

soft-constraintsarerepresentedasvariablesasinputfor

theoptimizationproblem.

Asimplematchingofpreferencevaluesrarelyleadsto

ane ectivetrade-o management.Thisisbecausepref-erencesgenerallydistinguishbetweenhardandsoftcon-straints.Hardconstraintshavetobeadheredto,nomat-terhow(un-)importanttherespectivepreferenceis.Con-sidertranscodinghints,wheretheoriginalauthorofthemultimediamaterialcande nehowthepropertiesofthemultimediacontentcanbechangedwithoutcompromis-ingthecontent’ssemantics.Forinstance,itcouldbestatedthattheresolutionofamoviecanonlybereducedupto50%oftheoriginalresolution.Afurtherreductionsimplydoesnotmakesense;evenifitisexactlythecon-tentauserrequestedbyexpressingcontentpreferenceswithhighpreferencevalues.Ontheotherhandausermightexpressapreferenceforbestpossibleresolution.Suchapreferencecanbeconsideredasasoftconstraintthatcanberelaxed,ifnecessary.Generallyspeakingtwohardboundariesalwaysconstrainservicecompositions(cf. gure2):thedevicecapabil-itiesformupperboundariesforthecapabilitiesofthe

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

4terminal,likeforexamplethemaximumpossiblereso-lution.Thetranscodinghintsformlowerboundariesfor

thequalityofmultimediamaterial,likeforexampletheminimumpossibleresolution.Ifthedevicecapabilitiesandtranscodinghintsdonotoverlapasensiblemediaadaptationwithoutchangingthemodalityisnotpossi-ble(cf. gure

2).

Fig.2BoundariesforMediaAdaptation

BasicallyforadaptationofthemultimediamaterialourSIAMserviceusestheInfoPyramidapproach[13]too ermultimediacontentinalternative delitiesand/ormodalities.Incaseofimpossibleadaptationstheuser’scontentpreferencesareusedtochoosealternativemediaandthusrealizeatrulycooperativeretrievalbehavior.

3RelatedWork

Thebasicusefulnessoftheuserpreferencesdescriptionschemeforautomaticcontentadaptationhasalreadybeenshowntosomedegreebyavarietyofapproaches.[25]providesanusefuloverviewofcurrentdigitalitemadaptationtechniquesinMPEG-21andhowtheycanbeutilizedbymultimediaapplications.Inparticular,theauthorsgivedetailedinsightsaboutusageenvironmentdescriptionswhichincludes,amongotherthings,termi-nalcapabilitiesandnetworkcharacteristics.FurthermoreongoingstandardizationactivitiesrelatedtoDIA,aswellasseveralemergingresearchtopicsandopenissuesarediscussed.

Thesemanticinformationamultimediadocumentcon-veyscanbepresentedwithindi erentlevelsofabstrac-tion.Recently[19]describedhowtoincludesemanticdescriptionsusingsemanticWebtechnology.ButsincetodaysemanticWeblanguagesstilllackthestructuraladvantagesoftheXML-basedapproachandcurrentmul-timediadocumentannotationisalreadygivenbyotherlegacystandards,acombiningoftheexistingstandardsseemstobethemostpromisingpathformultimediadoc-umentdescription.MPEG-7/21asoneoftheinterna-tionalkeystandardsinthisareaisde nedbyaXMLschemaandcurrentlynocommonlyacceptedmappingexistsfromtheXMLschemade nitionstoRDForOWL[19].Althoughafurthermonitoringofthedevelopment

BenjaminK¨ohncke,Wolf-TiloBalke

ofsemanticWebmediadescriptionsisnecessary,cur-rentlywestillrelyonXMLannotations.

Intermsofadaptationframeworks[20]alsousesthesim-pleXMLmodeltodesignavideopersonalizationandsummarizationsysteminheterogeneoususageenviron-ments.Thedesignframeworkresultsinathree-tierar-chitectureofserver,middlewareandclient.Personaliza-tionandadaptationengines,whichselect,adaptandde-liversummarizedmedia lestotheuser,areincludedinthemiddlewarelayer.Foreachdecisionifavideoshotshouldbeincludedinthepersonalizedvideosummaryornot,thepreferencescoresfromtheMPEG-7/21stan-dardaretaken.However,unlikeinourworkpresentedhereinallcasesthesepreferencesarestaticandtheap-propriateadaptationisalwaysassumedtobepossible.Incontrast,theworkin[21]goesbeyondmerestaticadaptationandevenprovidessomenovelontology-basedmethodologiestoopenuptheMPEG-7/21usageenvi-ronmentforenrichinguserpreferencesbymorecomplexsemanticsasexpressedbydomainontologies.Indoingso,theapproachsupportsthecompletefunctionalityo eredbytheMPEG-7semanticdescriptionschemeformulti-mediacontentdescriptionsandrespectsalltheMPEG-7/21conventions.ItisbasedontheWebontologylan-guageOWLandhasbeenprototypicallyimplementedintheDS-MIRFframework.Moreover,in[22]thesystemisextendedbyaQuerylanguageforMPEG-7descrip-tions.AllowingtheintegrationofdomainknowledgeintosemanticMPEG-7metadataanautomaticrelaxationofover-speci edconstraintsisthusenabled.However,theactualintegrationofspeci cuserpreferencesisnotyetsupportedinthisframeworkandthetaskofpersonaliza-tionstillremainsachallengingissue.

Inasimilarfashion[15]explorespersealgorithmsandmethodologiesforanontology-basedapproach,butherethefocusisonexchangingknowledgebetweenusers.Thesemanticrelationsintheontologyarede nedinMPEG-7andcanbevisualizedusingRDF.Anagglomerativeclusteringapproachisusedtoextractpreferencesbasedonuserhistoryandafuzzi edversionofthisknowledgecanbeexchanged.

Toexploitalsospeci cuserpreferencesandthereforeim-provetheuser’sinteractionexperienceandunderstand-ingoftheadaptationprocesstheintegrationofamoreexpressivepreferencemodelisneeded.Animportantsteptoderiveandmaintainuserpreferencesinameaningfulpro leistheelicitationstep.Thereexistsalargebodyofworkinusermodeling(seee.g.,[10],[5]forabasicoverview)relyingondi erenttechniquestoderiveapro- le.Especiallysuitablegrouppro lesascompromisesorabasisformulticasttechniquesmaybehelpfulinsolv-ingsomeofthedemandingQoSissuesinvolvedinmediastreaming.Forinstance,in[12]ausecaseofanelectronicprogramguideispresented,wheretheuser’sinteractivebehaviorwithvideoprogramsismonitored,andtheuserpreferencesareupdatedaccordingly.Here,therelativeimportanceofdi erentpreferencescantosomedegree

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

Preference-DrivenPersonalizationforFlexibleDigitalItemAdaptation5

evenbegeneratedautomatically.

Intermsoftheservice-orientedprocessingofmultime-diadatatheMPEG-21standardalsointroducesnetworkqualityofservice(QoS)tools.Forchoosingadequatenet-workpathstheusefulnessofintegratingmorecomplexuserpreferenceshasalreadybeenrecognizedandusu-allytypicalnon-functionalstatisticslikeaverageserviceavailabilityornetworklatenciesarealreadyintegratedinthedecision,e.g.fortelephony[8].However,theac-tualreal-timemeasurementofstatisticalinformationtodecideforbestservicepathsisstilldi cultandbeyondthescopeofthispaper(seee.g.,[7]or[16]).

4AMoreExpressivePreferenceModel

Theexistenceofmultipleandoftencon ictinguserpref-erencesdemandsane cientframeworktodealwiththeminafairandmeaningfulway.Theneedofane ectivetrade-o managementwithcomplexuserpreferenceshasalreadybeenconsideredinothercommunitieslikee.g.,indatabasesandinformationsystems.Here,recentworkin[6]and[9]considerspreferencesinaqualitativewayaspartialordersofpreferredvaluesthatcanberelaxediftheneedarises.TocombinemultiplepreferencesandderiveafaircompromiseusuallythenotionofParetooptimalityisused.

Inthecaseofcontentselectionandadaptationthere-sultsareallpossiblesolutionsthatarenotdominatedbyothersolutionswithrespecttoallattributesspeci edinthemediarequest.Thatmeansifallpreferencesondi er-entattributesareconsideredtobeofequalimportance,thesub-optimalsolutionsareautomaticallyremovedandoutoftheremainingpoolofpossiblesolutionsanadapta-tionservicecanpickasuitableinstantiation.Ofcourse,ifnofairrelaxationschemeisdesired,alsomorediscrim-inatingcombinationmethods(e.g.theorderingontheattributesinpreferencevaluesinMPEG-7/21)canbeusedonqualitativepartialorderpreferences.

Fig.3ExplicitPreferenceGraphs

Example(cont.):ConsiderouruserKimownsaPDAandagainrequestssomemediaforstreaming,butthistimeexplicitlyexpressestwocomplexpreferences:oneaboutpreferredmovieactorsandoneaboutpre-ferredcodecsavailableofherclientdevice(asde nedby

theterminalcapabilities).Thecorrespondingpreferencegraphsmightlookasshownin gure3.OurframeworkextendstheMPEG-21notationtoexpresstheseprefer-encesinXML,e.g.,the rstpreference,likefollows:

<UserPreference>

<UserIdentifieruserName="Kim"/><UsagePreference>

<Preference>

<EXPatt="actor"><EXPSet>

<Valueval1="Damon"val2="Willis">

<Valueval1="Damon"val2="Schwarzenegger"><Valueval1="Willis"val2="diCaprio">

<Valueval1="Schwarzenegger"val2="diCaprio"><Valueval1="Schwarzenegger"val2="Pitt"></EXPSet></EXP>

</Preference></UsagePreference></UserPreference>

Tocombineseveralpreferencesthecombinationse-manticshastobestated.Forourexampleletusassumeafairrelaxationschemebetweenthetwopreferences.Figure4showsthe rstthreelayersoftheproductpref-erencegraphfollowingtheParetosemantics.Thegraphisalreadyquitecomplexforcombiningonlythetwopref-erencesin gure3.Pleasenotethatduetothequalitativenatureofthepreferencessomecombinationsareincom-parable:ifthebestchoice(aMattDamonmovieinpxformat)shouldnotbeavailableoradaptable(e.g.,miss-ingasuitabletranscodingservice),theadaptationdeci-sioncanexploreseveralotheroptionsthatareallequallypreferablesuchasaMattDamon

movieinmpeg

formatoraBruceWillismovie

inpxformat.

Fig.4ParetoPreferenceGraph

ToexpresssuchParetopreferencesinXMLweallowqualifyingthecombinationofexplicitpreferencesfollow-ingtherespectivepreferencealgebra(see[9]fordetailsontheevaluationofcomplexconstraints):

<UserPreference>

<UserIdentifieruserName="Kim"/><UsagePreference>

<Preference><Pareto>

<EXPatt="actor"><EXPSet>

<Valueval1="Damon"val2="Willis">

<Valueval1="Damon"val2="Schwarzenegger"><Valueval1="Willis"val2="diCaprio">

<Valueval1="Schwarzenegger"val2="diCaprio"><Valueval1="Schwarzenegger"val2="Pitt"></EXPSet>

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

6BenjaminK¨ohncke,Wolf-TiloBalke

</EXP>

<EXPatt="codec"><EXPSet>

<Valueval1="px"val2="mpeg"><Valueval1="mpeg"val2="Huffyuv"><Valueval1="mpeg"val2="Indeo"></EXPSet></EXP></Pareto></Preference></UsagePreference></UserPreference>

Forevaluatingsuchanadvancedtrade-o manage-mentinaservice-orientedenvironmentseveralcompo-nentsareneeded.Figure5showsthebasicarchitectureofourpreferencetrade-o management.TheuserrequestsmediafromourcentralSIAMserviceandprovidescon-straintsandrichpreferencesintheformofcontentpref-erences(UP)andtechnicalcapabilities(TC).Thedeci-sionenginenegotiatesthecontentbyretrievingallnec-essarymetadataintegratingtherespectivetranscodinghints(TH)intothepreferencetrade-o

management.

initiatethechangetothecost-optimalsolution,whilereusingtheresultsofalreadyinvokedserviceswhennewbestservicechainsarederived.

E2MonisitselfrunasaWebserviceeitherontheclientor(moreoften)onaproxymachinetoreducethecom-municationloadoftheclientdeviceforWebservicein-vocationandmonitoring.Ofcourse,suchaproxy-basedapproachisonlyusefulifalarms(e.g.batterylow)arerarecomparedtoexternalevents(servicefailures),asotherwisecommunicatingthealarmscanleadtoahighcommunicationload.

5E cientlyEvaluatingPreferenceTrade-O sToe cientlyevaluatethecomplexmeshoftheParetoproductorderinducedbypreferencesspeciallyadaptedalgorithmsareneeded.Duetotheirpracticalusefulnessrecentlyinthe eldofdatabasessomeattentionhasbeenpaidtoconceptsforretrievingParetooptimalsets-so-calledskylinequeries,seee.g.,[4]or[18].

Skylinequeriesdescribethecasewhereallqueryat-tributesareconsideredtobeindependentandequallyimportant.Hence,noweightingfunctioncombiningin-pidualattributescoreslikeinconventionalintop-kre-trieval,canbeused.Instead,allpossiblyoptimalobjects,basedonthenotionofParetooptimality,arereturnedtotheuser.Anobjectisconsideredtobeoptimalwithre-specttoacollectionofobjectsandasetofpreferences,ifitisnotdominatedbyanyotherobject.Anobjectdom-inatesanotherobject,ifitshowsmoreoratleastequallypreferredattributevaluesasthisobjectwithrespecttoallattributesandisstrictlypreferredinatleastoneat-tribute.However,skylinequerieshavebeende nedwithtwoimportantconstraints:thedatahastobetotallyorderedforeachattributeandforeasycomparisonsat-tributeshavetobede nedovernumericaldomains.Withinskylineframeworksusersarealsoo eredthepos-sibilitytodeclareseveralhardconstraintsonattributes,asgivene.g.,bytranscodinghintsorterminalcapabili-ties.Thisisusuallyfacilitatedasasimpleselectioncondi-tionfor ltering[4].Butfortheuseinadaptationframe-workshardconstraintshavetobefurtherdistinguished.Thisisbecausesomehardconstraintscanstillbemetbyadaptingcontent(likeacodecorresolutionconstraint),whereasotherscanneverbeamendedbyalteringamediaobject(mostlyusagepreferenceslikepreferred/dislikedactorsorgenres).Wecallhardconstraintsofthe rsttypeadaptation-sensitivehardconstraints,whereaswerefertothesecondtypeasstricthardconstraints.

Werealizedtheactualexpressionofexplicithardcon-straintsbypreferencetagscontaininganadditionalat-tribute’constraint=”hard”’,whichsignalizetheSIAMservice(andallsubsequentWebservices)thatthispref-erencemustnotbeviolated.

Sinceinourframeworkallpreferencesarebasedonpar-tialorders,standardskylinequeryevaluationisthusnot

Fig.5Service-orientedAdaptation

ThedecisionenginepartoftheSIAMservicedecidesbymeansofthegivenconstraintswhichWebservicesarebasicallyneededtoperformanadequatemediadelivery.PossibleWebservices(whicharediscoveredbytheser-viceinstantiationpart)havetorespectthepreferences(andQoSconstraints)tochooseanoptimaladaptationofthemultimediadataforeachinpidualuser.

TheactualinstantiationandmonitoringenginesusedinourSIAMarchitecturearenotthefocusofthispaperandhavebeenextensivelydiscussedin[1].Inanutshell,theE2Mon(ExecutionandEnvironmentMonitoring)frameworkallowsafastselectionofnewservicechainsandaqualityandcostbasedservicefailurerecovery.To ndanoptimalservicechainacostfunctionisusedthatconsiderstheparametersfromtheactualexecution(suc-cessorfailure)andfurthermoreavarietyofcostsde-pendingondevicecapabilitiesortheexecutionenviron-ment.Ifthecostsofalreadydiscoveredserviceschange(e.g.,theexpectedQoS),eitheraprovideradvertisesthechangeinaserviceortheenvironmenthaschanged(e.g.,roamingtoanewnetworkoralowbatterywarning).Ineithercaseaneventisraisedandarediscoverywould

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

Preference-DrivenPersonalizationforFlexibleDigitalItemAdaptation7

readilyapplicable.Thoughitispossibletorenderatotalorderfrompartialorderpreferencesbyconsideringthe’level’ofobjectsineachpreferencegraph(thelevelbe-ingde nedbythelongestpathtoanymaximumvalueinthegraph),bydoingsoweacceptsomeinaccuraciesinducedbyobjectincomparability.

Example(cont.):Forinstance,takingtheactorpreferenceof gure3wecouldtransformthepreferenceintoatotalorderwith’Damon’asmostimportantvalue(onlevel1),followedby’Willis’and’Schwarzenegger’onthesameimportancelevel2,and nallyhave’Pitt’and’diCaprio’asleastpreferredvaluesagainwithsameimportanceonlevel3.Butwewouldalsostateinthetotalorderrenderedfromlevelinformationthat’Willis’ispreferredover’Pitt’thoughinfacttheyareincompa-rableinthepartialorderpreference.

Butfromthistotalorderinducedbylevelsitissim-pletoderiveanumericalassignmentofscorevaluesforobjectsbyusingautilityfunction,translatingthelevelinformationintosimplescores.Wenormalizethescoretoanumericalvaluebetween0(leastpreferred)and1(mostpreferred).ForeachrespectiveattributevalueainthedatabaseascoreSthequeryexpressedbyA(a)withrespecttoattributeAintotalorderpreferencepcanbecomputed,asfollows:

SA(a):=(maxlevel+1 actual)/(maxlevel)withmaxasthetotalnumberoflevelsinpreferencepandactualasthelevelonwhichattributevalueaactuallyoccurswithrespecttoAor(max+1),ifthevalueaisnotexplicitlystatedinp.

Indeo

diCaprio

Schwarzenegger

Fig.6ExampleSkylineforthePreferencesofFigure3

Example(cont.):AssumeouruserKimrequestsamovietobedeliveredtoherPDAandstatestheprefer-encesin gure3.Furtherassumethatthedatabasecon-tainsonlythefollowingobjects:a’MattDamon’movieinmpegformat(movie1),an’ArnoldSchwarzenegger’movieinmpegformat(movie2),a’BruceWillis’movieinpxformat(movie3),a’LeonardodiCaprio’movie

inHu yuvformat(movie4)anda’BradPitt’movieinmpegformat(movie5).Figure6showstheresultingtuples.Theskylineconsistsofonlytwoobjects(darkshaded):movie1andmovie3.Anadaptationdecisionengineneverneedstoconsideranyotherobjectfromthedatabase,i.e.movies2,4,and5(lightshaded),becausethosearede nitelydominatedbythetwoskylineobjects.Forexample,a’BradPitt’moviewillneverbeneeded(notevenasabaseforsubsequentadaptation),becauseitwillalwaysbedominatedbytheexisting’BruceWillis’movieinpxformatwithrespecttobothdimensions.However,pleasenotethataccordingtotheuserpref-erencesthemostpreferredobjectwouldbea’MattDa-mon’movieinpxformat(level0oftheParetograph).Buttheresimplyisnoinstanceinourdatabasethatsatis esthispreference.Inclassicalinformationsystemstheuserquerythusde nitelyhastoberelaxedtotheexistingskylineobjects.Ontheotherhand,inmulti-mediasystemsitisoftenpossibletoderiveanoptimalobjectbysuitablemeans.Thoughitisforinstancenotpossibleinourexampletocreatethemostpreferredob-jectbyadaptingtheWillis/px,moviebecauseweob-viouslycannotreplacetheactors,itisindeedpossibletotranscodetheDamon/mpegmovieintoaDamon/pxmovie(dottedarrowin gure6),ifthereisasuitabletranscodingserviceavailablethatmeetsallhardcon-straintscheckedbytheinstantiationengine(e.g.,QoSconstraints).Letussummarizesomedi erentpossibili-ties:

–RetrievetheMattDamonmovieinmpegformatandtranscodeitintopxformatbeforedelivery.–DelivertheBruceWillismovieinpxformat.–DelivertheMattDamonmovieinmpegformat.–RetrievetheArnoldSchwarzeneggermovieinmpegformatandtranscodeitintopxformatbeforede-livery.

–DelivertheArnoldSchwarzeneggermovieinmpegformat–...

Nowwewilltakeacloserlookofhowtocomputearank-ingforthebestadaptationdecisionbasedonthemoviesavailableinthedatabase.OfcoursethisrankingwillalsohavetobeaccordingtothelevelsoftheParetograph.Ourschemedescribedin[3]presentsafairrelaxationframeworkthatcanhandlesuchproblems.Theframe-workdescribeshowtorelaxsoftconstraintsgivenbytheuserifnoexactmatchshouldbeavailable.Firstthede-cisionenginehastobuildahierarchy(representingtherelaxationpath)outofallgivenpreferencegraphslikeshownin gure7.

Theavailabletuplesinourdatabaseareinstancesofthecorrespondingnodes(movie1,movie2,etc.).Thebestmatchingmovieful llsallsoftconstraintswithaminimumamountofrelaxationregardingasuitablequal-itymeasure.Sincethedi erentnodesinthepreferencegraphsarealreadyorderedaccordingtothenecessary

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

8

movie 1movie 5movie 2

movie 3

Fig.7PreferencesRelaxationPaths

relaxationstepsfromthetoporthebottomofthehier-archyorsimplytherelativedistancetotherootnodes.Wecanuseasimpleweightingfactortoalledgesinourhierarchygraphstodecidewhichmovietodeliver.Ourbasicassumptionforthecomputationoftheweightingfactoris,theclosewegettotherootofarelaxationpath(i.e.themorerelaxationisneeded),themoreunsatis edtheuserwillbewiththedeliveredresult.Wethusneedtoderiveaqualitymeasuretocomparebetweenmediaobjects.

Computationoftheweightingfactor(cf.[3]):Assumenpreferencegraphswiththeirrespectiverelax-ationhierarchies.Everyedgeislabeledbyn(d 1),wherenisthenumberofpreferencegraphsanddistherela-tivedistancetotherootnodetoachieveafair,breadth rstrelaxationscheme.Noweachdatabaseobjectoislabeledwithweightingfactorwf(o)bysummingupthelabelsofedgesrelaxedto ndthisobjectineachhierar-chy.Sortingtheobjectlabelsinascendingorderleadstoapreferenceorderofmediadeliverydecisions.

Example(cont.):Inourexamplemovie1andmovie2bothgotthesameweightingof1(cf. gure7).Iftheobjectshavethesameweightingsandstillnodecisionispossible,wehavearandomchoicebetweenthem.Thus,aimingataminimumnecessaryamountofrelaxationus-ingonlytheexistingmediaaccordingtoweareleftwithtwochoices:deliveringmovie1(aDamon/mpegmovie)needstorelaxtheconstraintonthecodecbyonelevelanddeliveringmovie3(aWillis/pxmovie)needstorelaxtheconstraintontheactorbyonelevel.

Beforecreatingthe naldeliveryorderwehavetotakealookifitispossibletoadaptoneoftheskylinemoviestobecomeabetterweightingfactor,respectivelyamoreoptimalsolution.Thepotentialadaptabilityisa-prioryknownforeachattribute.Weknowthatitisimpossi-bletoadapttheactorpreference( gure7,left)butitispossibletochangethecodecpreference( gure7,right)byasuitabletranscodingservice.Pleasenotethatthedecisionenginedoesnotneedtoknow,ifsuchaserviceisactuallyavailable,becauseitonlybuildsahierarchyof

BenjaminK¨ohncke,Wolf-TiloBalke

ingbacktoourexample,weseethatitisnotpossibletoimprovetheweightingfactorformovie3,butitispossibletoadaptmovie1bytranscoding.Withasuitabletranscodingser-vicetheweightingfactorofmovie1*(Damon/px)willbe0andbecomesindeedaninstanceoftheoptimalso-lutionofourpreferenceproblem(cf. gure6).

Pleasenotethatassoonastherearemediaobjectsalreadysuitableforimmediatedeliveryinthelist,nomorelessimportantobjects(i.e.notskylineobjects)thathavetobecreatedbyadaptationareneeded.Hencenotalldatabaseobjectshavetoberanked,butreallyonlyskylineobjectsorobjectscreatedbysuitableadapta-tionfromskylineobjects.Thespaceofpossibleadapta-tionsisthereforelimited.Thefollowingalgorithmgivesanoverviewaboutthewholeprocedure:

1./*Extractpreferencesaccordingtotheirattributes*/GivenisthesetPofpreferencesonnattributesinXMLformatbytheuserinteractiontools.

1.1.Extractthehardconstraintsforeachattribute

HC1.2.Divide1,...,HCHCn

constraints1,...,HCHCninadaptation-sensitivehard

1,...,HCjandthestricthardcon-straintsHCj+1,...,HC1.3.Extractthesoftconstraintsn

foreachattribute

SC2./*Transform1,...,SCpartialn

ordersSCi(1≤i≤n)intototalordersaccordingtotherespectivelevelinformation.Thisstepresultsinthepreferencerelaxationgraphs*/

Foreachcategoryi(1≤i≤n)do

putemaxlevelforSCpath

iaslengthoflongest

puteactuallevelforaasido

themaximum

puteS):=

i

(maxA(a+1 actuallevel)/(max)

2.3.BuildtotalorderSCi

usingtherankinginforma-tiongivenbyS3./*DeterminepossibleA(a)

mediaobjectsetbyskylinecom-putationandcomputerespectiverelaxationweights*/

GetinitialsetOofavailablemediaobjectsfromthedatabaseusingaskylinealgorithm(e.g.from[4])

withSCi

(1≤i≤n)asinputpreferencesand l-teredbystricthardconstraintsHC3.1.IfO= thenterminatewitherror

j+1,...,HCn/*Nosuitable(oratleastadaptable)mediaobjectisavailableindatabase*/3.2.Foreachobjecto∈Odo

puteweightingfactorwf(o)accordingto

preferencerelaxationgraphsSCi

(1≤i≤n)

3.3.Setdecision:=false

/*noadaptationdecisiontakenyet*/

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

Preference-DrivenPersonalizationforFlexibleDigitalItemAdaptation9

4./*Derivebestadaptationdecision*/While(notdecision)do

4.1.IfO= thenterminatewitherror

/*Nomediaobjectadaptationmeetshardcon-straints*/

4.2.ChoosesetD Oofobjectswithcurrentlylowest

wf(o)(o∈D)asbestavailableobjects

4.3.Ifwf(o)=0( o∈D)thencheckadaptation-sensitivehardconstraintsHC/*Thereisatleastoneperfect1,...,HCsolutionj

*/

4.3.1.IfHC1,...,HCjaresatis edforsomeo∈D

thensetdecision:=trueandreturnanysuchobjectoasadaptationdecision/*successfultermination*/

4.3.2.ElsediscardsetDfromOandifObecomes

emptyterminatewitherror

/*Nomediaobjectisadaptabletomeethardconstraints*/

4.4./*trytoadaptallobjectswithlowestwf(o)to

obtainsomeobjecto’withanevenlowerwf(o’)*/

Foreacho∈Dwithwf(o)>0(o∈D)do

4.4.1.Ifosatis esHC1,...,HCjthenputointoset

D ofpossibleadaptationdecisions

4.4.2.Tryallpossibleadaptationsofoasgivenby

SIAMwork ows[1]andSC

togetob-jecto withwf(o

)<wf(o)

1,,SCn

4.4.2.1.IfHC1,...,HCjaresatis edforo then

puto intosetD ofpossibleadaptationdecisions,elsediscardo

4.4.3.DiscardofromDandO

4.5.IfD = getobjectxfromD withminimum

wf(x),setdecision:=trueandreturnxasbestmatchingadaptationdecision/*successfultermination*/Thedecisionenginecannowderivethe naladapta-tionorderbysortingtheskylineobjectsandtheirpos-sibleadaptationsasoutputbythealgorithmaccordingtotheirrespectiveweightingfactor.Itthenhandstherespectivelistovertotheserviceinstantiationengine.Thisenginetakesthesoleresponsibilityforactuallyse-lectingtheservicechainbyiteratingovertheadaptationlistandcheckinginpiduallywhetherasuitablechaincanbeinstantiatedatthecurrenttime.

However,oneproblemthatmayoccurwhenusingsky-linequeryprocessingisthatParetosetsareknowntogrowexponentiallyinsizewithincreasingnumberofin-dependentpreferences.Asaremedyin[2]weintroducedanapproachtosigni cantlyreducetheskylinesizebyus-ingso-calledprimecuts.Primecutsareinterestingsub-setsofthefullParetoskyline,whicharerepresentativeoftherespectiveskylineandgenerallyprovidegoodcom-promises.ThekeytoimprovedperformanceandreducedresultsetssizeshereistherelaxationofParetoseman-ticstotheconceptofweakParetodominance(see[2]fordetails).Thesetofallnon-weakly-dominatedobjectsisreferredtoasthe’restricted’skyline.

IncomparisontothefullParetoskylinesourevaluationoveradatabasewith25000itemsin[2]shows,thatbyus-ingrestrictedskylinesthenumberoftuplesonlyslightlyincreases,ifmorepreferencesarestated(see gure8).Moreover,evenforlargenumbersofpreferencesthere-strictedskylinecanbecomputedabouttwoordersofmagnitudefasterthantraditionalParetoskylines(see gure9).Wecanstatethattherestrictedskylineap-proachmakesevenhigh-dimensionalpreferencetrade-o s(asmaysometimesbenecessaryinpersonalizedme-diaadaptation)

practical.

Fig.8Preferencedimensionalityandskyline

size

Fig.9Preferencedimensionalityandruntimes

6AMediaStreamingTestCase

Asaproofofconceptweimplementedthekeyfeaturesofthedecisionengineinasmalltestbed(calledPUMA)basedonWebservices.Sincewewereinterestedinde-rivingsuitableadaptationsourimplementationfocusedspeci callyonthechangesinthemetadatadocumentsandwhatcouldbederivedas naladaptationdecisions.Wedidnotperformtheactualadaptationofthevideo lesinthetestbed.AllWebserviceshavebeendeveloped

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

inC#.Foruserinter-actiontheSIAM(serviceinstantiationandmonitoring)

serviceisaddressedbyasimplePHPinterface,whereuserscanplacetheirinitialrequestswithadequatepref-erences.Likediscussedinsection4,theSIAMservicerunstheE2Monalgorithmfrom[1].

Figure10showstheelementsofourbasictestbed.Theexperimentalscenariofocuseddi erenttypicalscenariosliketheretrievalofvideosfromalternativedatabases,themergingwithsubtitlesindi erentlanguages,aswellastranscodingandqualityadaptation.Theinitialqueryparameters,theuserpreferencesandterminalcapabili-tiesarewrittenintoaMPEG-21document,whichissenttothecentralSIAMserviceforfurtherprocessing.

Fig.10ExperimentalScenario

Figure11showsascreenshotofthePUMAtestbed.Attributevaluesfor(hard)contentpreferencesorprefer-enceinformationintermsofpro les(likethePDAtech-nicalpro le)canbesubmittedtothesystem.Objectsareselectedfromthedatabaseaccordinglyandthenpre-paredforfurtherprocessingwithanappropriatemeta-datadescription lestatingallnecessarydetails.Example(cont.):LetuscomebacktouserKimwhowantsamoviestreamedtoherPDAwithherpref-erencesasdiscussedbeforeandadditionallyrequestsacertainscalingaccordingtoherclientdevice’sdisplaysizespeci edinthePDApro le.

Asdiscussedinsection5theserviceinstantiationen-ginegetsthelistfromtheadaptationdecisionengineand

willchooseanadequatework owfromasetofpat-terns.It rsttriesthetranscodingofthe’MattDamon’movieintopxformattodeliverthebestpossiblesolu-tionaccordingtothepreferencesstatedbyKim.Todothisithastointegrateasuitabletranscodingserviceintothedeliverywork ow.Sinceinmostmultimediaappli-cationswork owscanbeanticipatedandarewellunder-stoodthistaskisperformedbycheckingifanadequatetranscodingserviceisavailablebyrequestingthefunc-tionalityusingaUDDIservicedirectory.Iftheengine ndsasuitableservice,itwilltrytoinstantiatethecor-

BenjaminK¨ohncke,Wolf-TiloBalke

Fig.11ScreenshotofthePUMAtestbed

respondingservicechain.

Byhandingoverthepreferenceconstraints(includingespeciallyallhardconstraints)thetranscodingserviceisnowenabledtocheck,ifitispossibletotranscodethemovieintopx,e.g.matchingstatisticsontheex-pectedtimeneededfortranscoding,etc.againsttheuser-providedconstraints.Theservicemonitoringenginemon-itorsthewholeadaptationprocess.Hence,ifthetranscod-ingservicedecidesthatatranscodingisnotpossibleorthetranscodingfails,e.g.becausetheserviceisnotlongeravailable,arespectivealarmisraisedtothemoni-toringengineasdescribedindetailin[1].Nowtheserviceinstantiationenginewilltryto ndanequivalentservicetorecoverfromfailureordecidethatanadaptationoftheDamon/mpegmovietoaDamon/pxmovieisgen-erallynotpossible.Inthiscaseitwilltrytoinstantiatethenextalternativefromtheadaptationlist.

Ifthetranscodingsucceeds,ascalingservice( gure9:qualityadaptationservice)isstillneeded,toadaptthecurrentresolutionofthemovietothehighestpossibledisplayresolutionofthePDA(see gure2).Thisserviceagaingetsallpreferencedataanddecideslocallyifitissensibletoscaledowntheresolution.Tomakeadecisionaboutthat,theWebserviceforexamplecomparestheinformationfromthetranscodinghintstotheterminalcapabilities.Ifamismatchoccurs,againanalarmhastoberaised.Pleasenote,thatthoughsomecheckscanalreadybeperformedbytheadaptationdecisionengine,theservicesalwaysmayhavetointegratelocalparam-etersintothedecisionwhethertoperformataskandwhatisthebestway.

Moreover,inatypicalwork owKimmayalsorequestsubtitlesforacertainmovie.ThesubtitlesthencanalsobesavedintotheMPEG-21metadatadocumentandhandedontoamergingservicetogetherwiththeorigi-nalmediafromthedatabase.Afterthemergingprocesstheserviceremovesthenowobsoletesubtitlesandthe

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

Preference-DrivenPersonalizationforFlexibleDigitalItemAdaptation11

languagepreferencesfromthemetadatadocumentoftheadaptedmovie.and nallytheadaptedmovieisdeliv-eredtouserKim.

Throughoutourexperimentsitprovedpossibletohan-dlethepreferencematchmakingandthuscreateady-namicservicecompositionbasedonthemetadatawiththeSIAMserviceascentralcoordinator.

7ConclusionsandFutureWork

Withtheincreasingneedofserviceorientedconceptsincomplexmultimediaapplicationsalsotheadaptationofcontentissplitintoseveraltasks.Dependingontheavailablemedia,thediscoveredservicesandallseman-tic,user-speci candtechnicalconstraints,theselectionofthebestpossiblework owneedstosolveacomplexpreferenceproblem.ThecurrentpreferencemodelintheMPEG-7/21standard,however,islimitedtothesim-plematchingofnumericalvaluesrepresentingtheim-portanceofeachconstraint.Henceforcomplexinstanti-ationsofwork owsinaservice-orientedadaptationen-gineamoresophisticatedpreferencemodelisneeded.Byintegratingamodelofpartialorderpreferencesandevaluatingcomplexpreferenceconstraintsusingthesky-linequeriesascurrentlyexploredbythedatabasecom-munity,ourframeworkisabletohandletrade-o sinamoremeaningfulway.Wehaveshownontheexam-pleofatypicalmediastreamingscenariothatanau-tomaticadaptationofthemultimediacontentispossi-blebymeansofstructuredmetadatainformationandhowtoe cientlyexplorethecomplexresultmeshesthatemergefromcombiningindependentcontentandadap-tationpreferences.Thesepreferencesarecollectedfromindependentsourcessuchasexplicituserpreferences,in-trinsicdomainpreferences,transcodinghints,ortechni-caldevicecapabilities.Thiscompletepreferenceinfor-mationcanthenbeaugmentedbyallWebservicesinthework owandthusallowsanintelligentandhighlypersonalizedadaptationprocess.Moreover,everysingleWebserviceisenabledtodecidehowtoadaptmultime-diamaterialbest.

Afterimplementingthekeycomponentsasaproofofconceptwearecurrentlyimplementingtheentireframe-workincludingtheactualadaptationofvideostreamstogetfurtherinsightsonthescalabilityofourapproach.TheMPEG-21standardalreadyprovidessomedescrip-tionschemestocontainnetworkQoSparameters,whichwillhaveastrongin uenceonthescalabilityofservice-orientedadaptationprocesses.Wewanttoexploreinhowfartheseparameterscanbemeasuredorestimatedaccuratelyenoughtoguaranteestableservicecomposi-tions(worklatency,maychangeveryquicklyandleadtotheconsid-erationofalternativeservicechains).

Anotheropenquestionis,iftheroutingpathshouldalsobedeterminedbypreferences.Generallyspeaking,rout-

ingmediacanbeperformedinacustom-madeoverlayallowingthedistributionofservicesalongtheroutingpath.Here,thee ectivemanagementoftrade-o sforallservicesinvolvedcanleadtoabetteroverallser-viceforendusers.Finally,weplantointegratehigherlevelsemanticdescriptionsintoourframework.RecentlytheWorldWideWebConsortium(W3C)startedaSe-manticWebtaskforce[26]toinvestigatemultimediaan-notations.TherespectivestandardizationactivitiesarepromisingtoovercometheproblemofmissingmappingfunctionsforMPEG-7/21.References

1.W.-T.lectionBalke,bileModelJ.forDiederich:MultimediaAQuality-ServiceandCost-basedSe-2.(ICWS),Environments.W.-T.Chicago,IL,positiononWebServicesinMo-CutsBalke,U.G¨untzer,USA,2006.W.Siberski:GettingPrimeDatabasefromSkylinesoverPartiallyOrderedDomains.In3.(BTWsystemsinBusiness,TechnologieandWebW.-T.ingInMultipleBalke,2007),ConceptualM.Wagner:Aachen,Germany,ViewsThrough2007.

forQueryingDi A,WorldWideWebConference(WWW),NewYork,S.erator.B¨orzs¨2004.Inonyi,InternationalD.Kossmann,ConferenceK.Stocker:onDataTheSkylineOp-5.(ICDE),PTextBrusilovsky,Heidelberg,Germany,2001.

EngineeringandHypermedia.AKobsa,KluwerJVassileva:AcademicAdaptivePublishers,Hyper-USA,6.1998.

J.InChomicki:ACMTransactionsPreferenceonFormulasDatabaseinSystemsRelational(TODS),Queries.Vol.7.28X.(4),ServiceGu,2003.

K.Nahrstedt,R.Chang,C.Ward:QoS-assured8.works,W.InCompositionIEEEICDCSin2003,ManagedProvidence,ServiceRI,OverlayUSA,-Preference-basedKellerer,M.Wagner,bileMultimediaSignaling.SessionManagementW.-T.Balke,InEuropeanforH.Transac-tionsIP-basedSchulzrinne:Mo-on9.Telecommunications,W.Vol.15(4),Wiley,2004.

tems.Kießling:InInt.FoundationsConf.onVeryofPreferencesLargeinDatabaseSys-10.HongA.Kobsa,Kong,W.China,Wahlster:2002.

Databases(VLDB),Usermodelsindialog11.Springer,AdaptationB.K¨ohncke,USA,W.-T.1990.systems.Balke:PersonalizedDigitalItemWorkshop12.ization(SMAP),oninSemanticService-OrientedEnvironments.InInt.Athens,MediaAdaptationandPersonal-PersonalizedH.-K.Lee,J.Nam,B.Bae,Greece,M.Kim,2006.

K.Kang,J.Preference.InContentsAH2002GuideWorkshopandBrowsingonPersonalizationbasedonKim:User13.FutureDescriptionC.-S.Li,TV,R.Malaga,Mohan,Spain,J.R.2002.

inSmith:MultimediaAcoustics,SpeechintheandInfoPyramid.SignalProcessing,A,Seattle,WA,malD.Mukherjee,1998.

E.Delfosse,J.-G.Kim,Y.Wong:Opti-workAdaptationDecision-TakingforTerminalandNet-15.media,Quality-of-Service.Vol.InIEEETransactionsonMulti-RelationsP.Mylonas,7(3),shopinMultimediaM.2005.

Wallace:UsingOntologiesandFuzzy(SMAP),onSemanticPersonalization.InInt.Work-Athens,Greece,MediaAdaptation2006.

andPersonalization

Abstract The delivery of multimedia content often needs the adaptation of the content in order to satisfy user constraints. With the Digital Item Adaptation part, the MPEG-21 standard already defines a useful frame-work to handle this task. However, in mod

1216.ServiceK.Nahrstedt,putingProxyNetwork.D.Xu:FindingInSPIE/ACMServiceMultimediaPathsinaMedia

17.CA,andNetworkingConference(MMCN),SanCom-Jose,A,Nahrstedt,2002.

W.-T.Balke:TowardsBuildingLargeandStatus.MultimediaInACMSystemsWorkshopandApplications:position(MSC),Singapore,2005.

progressiveD.Papadias,algorithmY.Tao,forG.Fu,skylineB.Seeger:queries.AnInoptimalACMSIG-and19.MODMultimediaG.Stamou,ConferenceAnnotationsJ.van(SIGMODOssenbruggen,03),SanontheSemanticJ.Z.Diego,Pan,A,Schreiber:2003.InIEEE20.Multimedia,MPEG-21B.L.Tseng,Vol.forC.-Y.13(1),PersonalizingLin,2006.

J.R.Smith:UsingMPEG-7and21.Vol.Video.InIEEEMultimedia,renceC.Tsinaraki,11(1),2004.

DescriptionsS.Christodoulakis:inMPEG-7/21.SemanticUserPrefe-22.ManagementSymposium(HDMS),Athens,InHellenisticGreece,2005.DataModelC.Tsinaraki,trievalandS.Christodoulakis:AUserPreferenceshoponandanQueryLanguagethatAllowSemanticRe-SemanticFilteringMediaofMultimediaAdaptationContent.andPersonalizationInInt.Work-23.(SMAP),Athens,Greece,2006.

Salembier,P.vanBeek,scriptionSchemesJ.Smith,A.B.(VersionT.Benitez,Walker:J.3.1,WorkingMPEG-7Heuer,J.Draft),MultimediaMartinez,ISO/IEC,De-P.24.2000.

UniversalA.Vetro:MultimediaMPEG-21DigitalAccess.ItemInIEEEAdaptation:Multimedia,EnablingVol.25.11viewA.(1),Vetro,2004.

ofC.Timmerer:DigitalItemAdaptation:Over-26.TransactionsStandardizationMultimediaWordWideonMultimedia,andResearchActivities.InIEEEAn-notationWebVol.Consortium7(3),2005.onthe(W3C):27.Web./2001/sw/BestPractices/MM/SemantictoM.mediaDynamicZhou,V.DataAggarwal:ContentAnOptimization-basedSelectionApproachSoftwareInterfaces.andTechnology,InACMSantaSymposiuminFe,NM,onIntelligentMulti-USA,

User2004.

InterfaceBenjaminK¨ohncke,Wolf-TiloBalke

BenjaminMasterK¨ohnckecurrentlyisnoverComputerandstudenthasatreceivedUniversityhisofaBAHan-ofHannover,ScienceGermanyatthein

2006.

UniversityinWolf-TilociatesearchresearchBalkesince2004asso-nover.CenterofdirectorofL3SRe-fellowatatBeforethatUniversityhewasaresearchofHan-ofGermantheBerkeley,theUniversityEmmy-Noether-ProgramUSA.HeisofaCaliforniamemberreceivedResearchahisMSinFoundation.HeofthehasversityPhDinofcomputermathematicsandAugsburg,scienceGermany.

fromUni-

本文来源：https://www.bwwdw.com/article/gahq.html