适性化多代理人网际网路环境资讯侦搜

適性化多代理人網際網路環境資訊偵搜CollaborativeMultiagentAdaptationforBusinessEnvironmentalScanningthroughtheInternet劉瑞瓏Rey-LongLiu中華大學資訊管理系中華民國92年11月18日2Outline•Introduction–BusinessEnvironmentalScanningthroughtheInternet(ES)–User-Centered,Continuous,andResource-BoundedES(UCRES):Goal&challenges–IntelligentMultiagentTechnology•MultiagentAdaptationforUCRES–OverviewofAESA–TheScanningAgents–TheControllingAgent•Experiment•Conclusion3Introduction•BusinessEnvironmentalScanningthroughtheInternet(ES)–Aroutineactivityforcollectinginformationofinterest(IOI)fromthe•Informationaboutvariousentitiessuchasthegovernment,competitors,customers,andpartners–Supportingthebusinessestorespondtotheneweststatusoftheenvironmentinatimelymanner4•UCRES:User-centered,Continuous,andResource-boundedES–Goal&clallenges•FindingIOIcontinuously–Timeliness(TES):minimizingtheaveragetimedelayoffindingIOI–Completeness(CES):maximizingthepercentageofIOIfound•Controllingtheresourceconsumed–Effectiveness(EES):maximizingthepossibilityoffindingIOIineachinquiryofdata•Consideringthepreferencesoftheuser(i.e.addingimportanceweightstoTES,CES,andEES)–WeightedTES–WeightedCES–WeightedEES5•RelatedmultiagenttechnologyforUCRES–Informationgatheringagents•Aimingtosatisfyusers'one-shotneeds,e.g.–Intelligentlylocatingtheinformationwiththeadditionalconsiderationoftime/cost/qualitytradeoffs,and–Filteroutirrelevantinformation•ButwhenandhowfrequentlytoscanforIOI?–Informationmonitoringagents•Aimingtomonitorapredefinedsetoftargetsperiodicallyoradaptively•ButwheretofindtheIOItomonitor?6–Adaptiveagentsfordecisionsupport•Aimingtoprovidetailoredinformationforsupportingdecisionmaking•Adaptingtowhattomonitorbyobservingtheuser’spreferenceand/orproblemsolvingstrategies•ButwhenandhowfrequentlytoscanforIOI?–Multiagentcoordination&bidding•Aimingtoresolveconflictsandbuildconsensusamongtheagents•Buthowtodefineacoordinationand/orbiddingprotocolforUCRES?7–Learningformultiagentcoordination,bidding,andorganization•Aimingtolearn–Howtheactionsaffecteachother,–Whatinformationisrequiredforcoordination,–Whentotriggeragentcoordination,–Usageofwhattheagentsarebiddingfor,–Restructuringoftheorganizationforcoordinationandcollaboration•Buthowtosimultaneouslyadaptto–User’spreference–IOI’sdistributionintheInternet–UpdatebehavioroftheIOI–Limitedamountofresource8MultiagentAdaptationforUCRES•AESA:AdaptiveESAgents–Ascalablesocietyofautonomousagents•Asetofscanningagents•Acontrollingagent–Basicrulesoftheagentsociety•Collaboratingforsatisfyingtheuser’sinformationneeds•Sharingthelimitedamountofresource–Adaptationto…•User’spreference•IOI’sdistributionintheInternet•UpdatebehavioroftheIOI•Limitedamountofresource9OverviewofAESASystemAdministratorNotificationEvaluateScanningAgent3ControllingAgentTerminateAdjustResourceAESASpawnScanningAgent1ScanningAgent2ScanningAgentnInformationPreferencesResourceLimitUser1User2UsermMonitoringDiscoveryIntranetThe(1)Goal=Setofinformationpreferencestobesatisfiedbytheagent;(2)Target=TargetwebsitefromwhichtheagentisdelegatedtofindIOI;(3)ResourceU=Upperboundofresourceconsumptionbytheagent;Repeat(4)Iftheagentisreadytoretrieveinformation(accordingtoResourceU),(4.1)RetrieveinformationfromTarget;(4.2)Ifnewinformationisfound,(4.2.1)IssueaType-Erequesttoget(DegreeOfSatisfaction,SatisfiedGoal);(4.2.2)Aliveness=Aliveness*+DegreeOfSatisfaction;(4.2.3)Triggersuitableprocedures,includingeventloggingandusernotification;(4.2.4)IfAliveness,foreachtargetdfoundinthenewinformation,(4.2.4.1)IssueaType-Srequesttospawnanewagent(4.3)Iftheagentsucceedstwotimes,issueaType-R-Morerequestformoreresource;(4.4)Iftheagentfailstwotimes,issueaType-R-Lessrequestforreleasingresource;(5)Ifaninformationpreferenceisnolongervalid,removetheneedfromGoal;UntilAlivenessorGoalisempty;(6)TerminatetheagentbyissuingaType-Trequest.11BehavioroftheControllingAgent(1)=Upperboundoftheresourcethatmaybeconsumedbythesystem;Repeat(2)Ifauserentersanewpreferencee,generateanagent(ifnecessaryandpossible);(3)Ifauserremovesaninformationpreferencee,informallagentstoremovee;(4)IfthereisaType-R-Lessrequestfromagentk,(4.1)Release(fromk)1percentsofresource;(5)IfthereaType-R-Morerequestfromagentk,(5.1)Allocate(tok)2percentsofadditionalresource(byagentaliveness);(6)IfthereisaType-Erequestfromagentk,(6.1)ReplykwithDegreeOfSatisfaction=(DOSe*Importancee),foreachgoale;(7)IfthereisaType-Srequesttospawnanewagent,(7.1)Replytherequestingagentwith(ifpossible);(8)IfthereisaType-Trequestfromagentk,terminatekandreleaseallresourceofk;(9)Ifthereisnoresponsefromkforalongtime,terminatekandreleaseallresourceofk;Untilthesystemisterminated.12ExperimentThreedimensionsoftheupdatebehaviorofawebsitewwhosecentralcategoryisc:(1)Whentoupdate:followingexponentialupdateTheprobabilityofupdatingwintimeintervalxfollowstheprobabilitydensityfunctionf(x)=e-x,whereistheaverageupdatefrequencyofw.(2)Whichparttoupdate:(2a)Maincontentisupdatedwithaprobabilityof0.5.(2b)Embeddedhyperlinksareupdatedwithaprobabilityof0.5.(3)Waytoupdate:preservingi