ExploringtheRelationshipsamongICTs:AScalableComputationalApproachUsingKLDivergenceandHierarchicalClusteringChia-jungTsui,PingWang,KennethR.
Fleischmann,DouglasW.
Oard,andAsadB.
SayeedUniversityofMaryland,CollegePark{ctsui,pwang,kfleisch,oard,asayeed}@umd.
eduAbstractDifferentinformationandcommunicationtechnologies(ICTs)arerelatedincomplexwaysand,accordingly,theirdiffusiontrajectoriesarerelated,too.
HowcantherelationshipsamongmultipleICTsbedescribedandanalyzedinascalablewayInthisstudy,weofferascalablemethodology,basedoncomputationalanalysisofdiscourse,toexaminetherelationshipsamongICTs.
Specifically,weemployedKullback-Leibler(KL)divergencetocomparethesemanticsimilarityofforty-sevenICTsdiscussedinthetrademagazineInformationWeekoveradecade.
Usinghierarchicalclustering,wehavefoundthatthesimilarityofthetechnologiescanbemappedinahierarchyandsimilartechnologiesdemonstratedsimilardiscourses.
Theresultsestablishthevalidityofourapproachanddemonstrateitsscalabilityandrichness.
Thisanalyticalapproachnotonlyenablesdiffusionresearcherstoundertakemulti-innovation,multi-source,andmulti-periodstudies,butalsohelpspractitionerseffectivelyadoptandefficientlyusenewICTsintheirorganizations.
1.
IntroductionPractitionerswhoconsideradoptingandusingnewinformationandcommunicationstechnologies(ICTs)andscholarswhostudythediffusionofICTsfaceaconstantchallenge:Atanyonetime,weconfrontnumerousseeminglypromisingICTinnovations.
Someofthembecomewidelyadoptedandused,makingsignificantcontributionstoeconomicprosperityandsocialwelfare;whereasothersfadeaway,leavinglittletracebehind.
WhileithasbeenarguedthatvariousICTsarerelatedtovaryingdegreesandsoaretheirdiffusiontrajectories[19],itisdifficulttomakesenseoftherelationshipsamongICTinnovations.
Forexample,hereisapartiallistofcontemporaryICTinnovations:Service-orientedarchitecture(SOA),webservices,opensourcesoftware(OSS),web2.
0,YouTube,iPhone,blogs,andutilitycomputing.
HowaretheyrelatedHowaretheirdiffusiontrajectoriesrelatedThesearedifficultquestions,notonlybecauseofthecomplexrelationshipsamongICTinnovations,butalsoduetothelackofreliablemethodstodescribeandanalyzetherelationships.
Indeed,ICTinnovationsarerelatedincomplexways.
First,abroaderconceptmaybecomprisedofnarrower,morespecificconcepts.
Second,differentconceptsmayrepresentthesamecoreidea.
Third,conceptsmaycompetewitheachotherasalternativesolutionstosimilarproblemsorfortheattentionfromthesamegroupofpeopleororganizations.
Finally,conceptsmaycomplementeachothertoaccomplishcommontasks.
Overtime,theserelationshipsmaychange,makingitevenhardertointerpret.
ResearchersofICTdiffusionarenotwellequippedtodescribeandanalyzethecomplexandevolvingrelationshipsamongICTinnovations.
Ontheonehand,manystudiesinthedominantparadigmofICTdiffusionresearchhavedemonstratedthatvariousorganizational,technical,andenvironmentalfactorsinfluenceICTadoptionanduse[8].
Asthisdominantparadigmisreaching"thepointofdiminishingreturnsasaframeworkforsupportingground-breakingresearch"[8,p.
314],itshouldbenotedthatmoststudiesinthedominantparadigmemploysingle-innovationresearchdesigns,leavingtherelationshipsamongICTsunderexplored.
Ontheotherhand,thefewmulti-innovationstudieshavehadtoexplicitlyorimplicitlyrelyondomainexpertstoevaluateICTrelationships(e.
g.
,[6],[19]).
Suchexpertevaluationsaredifficulttoreplicate,togeneralizetootherICTs,ortoscaleuptoexaminetherelationshipsamongalargenumberofICTs.
Therefore,consideringthecurrentstatusoftheICTdiffusionliterature,weraisethisresearchquestion:HowcantherelationshipsamongalargenumberofICTsbedescribedandanalyzedinascalablewayWeanswerthisquestioninthisstudybyofferingascalablemethodology,basedoncomputationalanalysisofthediscourseaboutmultipleICTs,tounderstandinnovationrelationships.
Inthefollowing,wefirst1Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010978-0-7695-3869-3/10$26.
002010IEEEbrieflylaythetheoreticalfoundation.
Thenweillustrateourapproachwithanempiricalstudyof47ICTsoveradecade.
AndfinallyweconcludebydiscussingthevalidityandbenefitsofourapproachforICTdiffusionresearchandpractice.
2.
ArelationalviewofICTinnovationsAnICTinnovationisaninformationandcommunicationtechnologythatisperceivedasnewbyanindividualorotherunitofadoption[15].
Noinnovationemergesandevolvesinisolation.
Tovaryingdegrees,thediffusiontrajectoriesofrelatedICTinnovations(asmaybeindicatedbypopularityorcumulativeadoption)areinterrelated.
However,findingsfromthefewsporadicstudiesofthediffusionofrelatedinnovationsareinconsistent.
Ontheonehand,forexample,Wang[19]foundthatthepopularitywaves(asmeasuredbythenumberofarticlesabouteachinnovation)offourICTinnovations(manufacturingresourceplanning–MRPII,businessprocessreengineering–BPR,customerrelationshipmanagement–CRM,andsupplychainmanagement–SCM)thatwerecloselyrelatedtoenterpriseresourceplanning(ERP)werenegativelyrelatedtothepopularityofERP,indicatingcompetingorsubstitutingrelationshipsbetweenERPandtheotherinnovations.
Ontheotherhand,BergerandHeath[3]foundpositivecorrelationamongrelatedculturalinnovations,confirmingamainthesisinsocialcognitiontheory:Whenaparticularconceptisactivated,relatedconceptsmaybeactivatedaswell[9].
Toresolvethisinconsistency,wemustunderstandhowICTinnovationsarerelated.
OneusefulapproachistoclassifyICTinnovationsbytheirattributesandfunctions.
Forexample,Ein-DorandSegev[6]surveyedthedefinitionsof17ICTinnovationsintheInformationSystemsliterature,identifiedfromthedefinitions31attributesand27functions,andthendescribedtheinnovationsbytwobit-vectors:avectorofattributesandavectoroffunctions.
Furthertheyperformedquantitativemethodssuchasmultidimensionalscaling(MDS)tovisualizetherelationshipsamongtheinnovationsintermsoftheirrelativesimilarity/dissimilarity.
Whilethisapproachcanbeimplementedrigorouslywithinthescopeofeachstudy,thechoiceoffactors(suchasattributesandfunctionsofICTs)forclassificationvaries,dependingonthespecificopinionsorbackgroundknowledgeofthedomainexperts(mostoftentheauthorsthemselves)whoparticipateinthestudy.
Inaddition,asthenumberofICTinnovationsincreases,theeffortbyhumanexpertstodescribeeachinnovationaccordingtoitsattributesandfunctionsincreasesandthereliabilityofthatclassificationworkmaydecrease.
Insearchofanalternative,complementaryapproachtounderstandingICTinnovationrelationships,weproposetoconceptualizeanetworkofinnovationsaspartofanecologicalsystem,whereinnovationscanbelikenedtospeciesinacompetitiveandsymbioticresourcespace[19,21].
Innovationsrelyontheattentionfromcommunitiesoforganizationsandpeoplewithinterestsinproducingand/orusingtheinnovations.
Eachcommunityemergestomakesenseofaninnovationandorchestratematerialactivities.
Themembershipofthecommunityevolvesdynamically,asthecollectiveattentiontotheinnovationevolves.
Theflowsofattentionamonginnovationsarebothreflectedandenabledbydiscourse–whathavebeensaidandwrittenabouttheinnovations.
Thediscourseaboutaninnovationcandidlyrecordshumanactionsundertakentodescribe,promote,ordenouncetheinnovation[14].
Therefore,analysisofthediscourseaboutmultipleinnovationscanhelpusassessthecomplexrelationshipsamongICTinnovations.
Althoughdiscourseanalysisdoesnotnecessarilyrequirethehelpofdomainexperts,wherethereisalargeamountofdiscoursedata,researchersoftenhavetotradeoffbetweendepthandbreadthindiscourseanalysis.
3.
ComputationalapproachtotheanalysisofICTrelationshipsRecentadvancesincomputationallinguisticshavemadeitpossibletoachievebothdepthandbreadthintheanalysisofverylargesetsofdiscoursedata.
Computationalorautomatedanalysisofdiscourseisalarge,activeinterdisciplinaryfieldwithavarietyoftheoriesandtechniques(see[13]foranon-technicalprimer).
Inthispaper,weofferacomputationalmethodologythatcandescribeandanalyzeinnovationrelationshipsinascalableway.
Wehaveconductedanempiricalstudytoillustratethismethodology.
Beforewepresentthedetailsofourapproach,weneedtodescribethediscoursedatawehavecollectedforthisillustration.
3.
1.
DatacollectionTherearenumerousdiscourseoutlets,includingbooks,magazines,conferences,blogs,wikis,andmanyothers,wherediscoursedatamaybecollected.
InordertoillustratehowourscalablecomputationalmethodologyworksinICTinnovationresearch,wedecidedtofocusonaparticularICTtrademagazine,InformationWeek,asthedatasourceforthis2Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010illustration.
Asdescribedbelow,thescaleofthedatawehavecollectedfromInformationWeekislargeenoughforustodemonstratethescalabilityofourapproach.
Whilewecouldhaveselecteddataofasimilarorlargerscalefromanotheroutletorsetofoutlets,wechosetofocusonatrademagazinefortwomainreasons.
First,comparedtoothertypesofoutlets,thetradepressinvolvesmost,ifnotall,typesofactorsinaninnovationcommunity(e.
g.
,vendors,users,consultants,academics,regulators,investors,andjournalists)intheproduction(e.
g.
,writing,editing,andadvertising)andconsumption(e.
g.
,subscribing,reading,anddiscussing)oftradepublications.
Second,itisthebusinessofICTtrademagazinestoactivelyrepresentwhatothershavetosayaboutvariousICTinnovations[20].
Whilesometimesaparticulartrademagazinemaybeoverinfluencedbyaparticularactor(suchasaninfluentialvendororimportantusergroups),foranytradetotakeplace,thetradepressbydefinitionhastostrikeabalancebetweenthepursuitsofvendorsanduserneeds[20].
Suchcomprehensiveanddiverserepresentationoftheactorsandtheirideasandactionsmakestradepressadesirablesourceofdatainthisstudy.
Further,asoneofthemajorICTtrademagazines,InformationWeekreachesapproximately440,000businesstechnologyprofessionalsatmorethanaquartermillionuniquelocations.
Overnearlythreedecades,themagazinehasbeenhelpingITleadersdefineandframetheirbusinesstechnologystrategies.
Foranillustrativestudylikeours,InformationWeekprovidesanappropriatelyfocusedsourceofdata.
Wedownloadedallofthearticlespublishedduringaten-yearperiod(1998-2007)inInformationWeekusingtheLexis/Nexisonlinedatabase.
Meanwhile,wecompiledalistof47ICTinnovationconcepts(Table1),rangingfromenterprisesoftware(e.
g.
,CRM)topersonalgadgets(e.
g.
,iPod),fromabstractconcepts(e.
g.
,artificialintelligence)toconcreteproducts/services(e.
g.
,YouTube),andfromhighlypopular(e.
g.
,e-business)tolesswell-knownconcepts(e.
g.
,digitalsubscriberline–DSL).
Admittedly,thislistisadhoc,butitservestheillustrationpurposewellbecausethelistcoversabroadrangeofICTinnovationsintheexaminationperiod.
WethenextractedfromtheInformationWeekarticlesallparagraphsthatcontainanyoftheICTinnovationsonthelist.
Indoingso,weconsideredpossiblelabelsforeachinnovation,pluralforms,andacronymsuniquetotheinnovation.
Forexample,inextractingparagraphscontaining"digitalsubscriberline,"wealsoincludedparagraphsmentioning"digitalsubscriberlines"and"DSL.
"SomeICTinnovationshadmanyparagraphsinthe10-yearperiodwhileothershadonlyafew.
Forexample,thereweremorethan5,000paragraphsmentioningERP.
Intotal,71,113paragraphswereextracted,withabout1,500paragraphsonaverageforeachinnovation.
3.
2.
DataanalysisTomakesenseoftherelationshipsamongtheICTinnovations,wefocusedforthispaperontheinitialstepofexploringthesimilarityoftheinnovations.
Oneapproachistoinferinnovationsimilarityfromthesemanticsimilarityofthediscoursesabouttheinnovations.
Specifically,weemployedKullback-Leibler(KL)divergence,aprobabilisticmeasurefordifferencesinthepatternofwordchoicesbyauthors,asaproxyforcomparisonofthesemanticsimilarityofanytwoparagraphsextractedfromInformationWeek.
Wealsousedhierarchicalclusteringanalysistoaggregatetheinnovationsinahierarchicalstructure.
3.
2.
1.
KLdivergence.
Originallyintroducedin1951[11]andconsideredafoundationofinformationtheory[5],KLdivergenceisastatisticthatquantifiesinbitshowcloseaprobabilitydistributionPistoanotherdistributionQ.
Forprobabilitydistributionsofdiscreterandomvariables,theKLdivergenceofQfromPisdefinedas:log(DPQPiPiQiKLi∑=.
KLdivergenceisalwaysnon-negative.
Itequalszeroifandonlyifthetwodistributionsmatchexactly.
KLdivergenceiscommonlyusedforcomparingtherelativefrequencyoftermuseinpairsofdiscourses[12].
Inthisdataset,eachICTinnovationisrepresentedbyconcatenatingalloftheparagraphsthatwereautomaticallydetectedasmentioningtheinnovation.
TheuseoflanguageintheparagraphsconstitutesaprobabilitydistributionofnormalizedunigramwordcountsandwecalculatedtheKLdivergenceforeachpairofinnovations.
Thecalculationgeneratedanasymmetric47x47matrixwitheachcolumnandrowrepresentingoneofthe47innovations.
Aftersymmetrization(byaveragingtheKLdivergenceineachdirection,i.
e.
2KLKLDPQDQP+),thevalueineachcellofthematrixdefinesadistance(intheformalsense,satisfyingthetriangleinequality)betweenapairofinnovations.
Inordertoclassifytheinnovationsandvisualizetheirrelationships,weperformedclusteringanalysisonthesymmetrizedKLdivergencematrix.
3.
2.
2.
Hierarchicalclustering.
Clusteranalysisistheprocessofgroupingobjectsintounknownclusterssuchthatthewithin-groupvariationisminimizedandthebetween-groupvariationmaximized[7].
The3Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010agglomerativehierarchicalclusteringmethodgroupsobjectsonaseriesoflevels,fromthefinestpartition,inwhicheachindividualobjectformsitsowncluster,andsuccessivelycombinessmallerclustersintolargeronesuntilallobjectsareinonecluster.
Agglomerativehierarchicalclusteringemploysanaggregationcriterion,or"linkagerule,"todeterminehowthedistancebetweentwoclustersshouldbecalculatedbasedonthedistancescoresofpairsofobjects.
Themostwellknownaggregationcriteriaaresinglelink,completelink,andaveragelink[10].
Thedistancebetweentwoclustersisrepresentedbytheminimum,maximum,oraveragedistancebetweenanypairofobjects,oneobjectfromeachcluster.
Insinglelinkclustering,twoclusterswiththesmallestminimumpairwisedistancearemergedineachstep.
Incompletelinkclustering,twoclusterswiththesmallestmaximumpairwisedistancearemergedineachstep.
Andaveragelinkclusteringisacompromisebetweentheothertwomethods.
Weusedthecompletelinkinthisstudybecauseitformstightclusterswithgoodglobalclusterquality[12].
4.
ResultsOurclusteringanalysisgeneratedahierarchyofclustersinadendrogram(Figure1),whereverticallinesshowjoinedclustersandthepositionofthelinesonthescalefrom1to25indicatesthedistanceatwhichclustersaremerged.
Byinspectingthedendrogram,wehaveidentifiedfivenaturalclusters,allofwhichmergedbetween15and20inthehorizontalscale.
TheseclustersareindicatedbythefiveintersectionpointsbetweenthedendrogramandtheverticaldottedlineinFigure1.
Cluster1includes26ICTinnovations.
MostofthemareenterpriseICTapplications.
Thehierarchicalstructureofthislargeclusterisshowninthedendrogram.
Forexample,atthenextgranularlevel(around15inthehorizontalscale),thereexisttwosub-clusters:oneconsistingofservice-orientedICTinnovationssuchasOSSandwebservicesandtheotherrepresentingmoretraditionalICTs,whichmaybefurtherdifferentiatedatlowerlevels.
Withinthelattersub-cluster,wecanseethat,forexample,thediscourseone-businessisverysimilartothatone-commerce.
SimilarrelationshipsseemtoexistininnovationpairssuchasCRMandERP,andknowledgemanagement(KM)andgroupware.
WecountedthenumberofparagraphseachyearmentioningeachICTinnovation.
Thenumberofparagraphsaboutaninnovationindicatestheprevalenceorpopularityoftheinnovationinthediscourse.
Forexample,Figure2showsthatthepopularitycurvesofthepairofverysimilarinnovations(e-businessande-commerce)followedverysimilarpatterns:bothinnovationsenjoyedpeakpopularityaround2000andthenhavelostmuchmomentumsincethedot-comcrash.
Incontrast,thepopularitycurvesofotherverysimilarinnovationsfollowedverydifferentpatterns.
ConsiderFigure3,whichshowstheevolutionarytrajectoriesofwebservicesandSOA.
ThenegativelycorrelatedcurvesinthefigureseemtosuggestthatthenewerSOAreplacedtheolderwebservices.
Cluster2includesfiveICTinnovations:DSLandvirtualprivatenetwork(VPN)aretelecommunicationtechnologieswhichcanbeappliedtotheotherthreeinnovationsinthecluster.
AsFigure4shows,DSLandVPNhadverysimilarpopularitytrajectories.
OnlythreeinnovationsformCluster3andfourinnovationsformCluster5.
Thesetwoclusterscorrespondtotheso-calledweb2.
0technologiesthathavebecomehighlypopularinrecentyears.
InnovationsinCluster3sharesocialnetworkasacommonfeature.
Cluster5representstext-basedweb2.
0applicationswithusergeneratedcontents.
ThepopularitycurvesinFigures5and6showthattheinnovationsinthesetwoclustershavegenerallyexperienceddramaticupswingscirca2004.
Despitethesimilarity,thepatternsoftermuseinthetwoclusters(asmeasuredbysymmetrizedKLdivergence)donotconvergeatthenextlevelofaggregation.
Thisinterestingfindingseemstosuggestthesubstantialdiversityofweb2.
0technologies.
Lastly,Cluster4hasnineinnovationsallrelatedtomobileorwirelesstechnologies.
Some,suchasbluetoothandWi-Fi,aretheunderlyingmobiletechnologies.
Others,suchasTabletPCandiPod,arethedevicesenabledbythewireless/mobiletechnologies.
Figure7showsthattherisingpopularityofiPhonecoincidedwiththedwindlingpopularityofiPod,suggesting,onceagain,thatthenewreplacestheold.
5.
Discussion5.
1.
ValidityoftheapproachTheresultsfromtheKL-divergenceandclusteringanalysisareconsistentwithouraprioriknowledgeabouttherelationshipamongthese47ICTinnovations.
Suchconsistencyprovidesreasonableconfidenceinthevalidityofthestudy'scomputationalapproachtounderstandinginnovationrelationships.
Therefore,thisstudynotonlyillustrateshowourmethodologyworks,butalsovalidatesourmethodology,whichmaythenbeconfidentlyappliedtothemorecommonscenarios4Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010whereaprioriknowledgeisunavailable,suchasthecasesofneworunknowninnovations.
Further,wehavetriangulatedtherelationshipswediscoveredinthisstudybyemployingmultidimensionalscaling(MDS),anothervisualizationtechnique,toreducetheKLdivergencematrixtotwodimensions.
Baseduponamatrixofitem-itemsimilaritiesordissimilarities,anMDSalgorithmassignsalocationtoeachiteminaspacesuchthatthedistancesbetweentheitemscorrespondascloselyaspossibletothemeasureddissimilaritiesbetweentheitems.
Inotherwords,theproximityofitemstoeachotherinthespaceindicateshowsimilartheyare.
WeusedtheMDSprocedureinSPSSbasedontheALSCALoralternatingleastsquaresscaling[17],themostpopularMDSalgorithm.
InFigure8,wepresentthe47ICTinnovationsintheresultingtwo-dimensionalscatterplot.
ThisplothasanR-squaredof0.
72,meaningthat72%ofthevarianceofthescaleddatacanbeaccountedforbytheMDSprocedure.
Inthisfigure,wehaveuseddifferentcolorstocorrespondtothefiveclustersidentifiedinthepreviousclusteringanalysis.
Generallyspeaking,mostoftheinnovationsinthesamecluster(asshowninthedendrograminFigure1)arelocatedclosetoeachotherintheMDSplot.
Thisadditionalanalysishasstrengthenedourconfidencethatthis"KLdivergenceplushierarchicalclustering"analyticalapproachyieldsresultsthatcanbeinterpretedfairlyeasily,andthatcomportwithourintuition.
5.
2.
BenefitsoftheapproachOurapproachhasseveraladvantages.
Foremost,computationalanalysisisscalable.
Thestudyhasexaminedthediscourseof47innovationsintenyears,alreadysurpassingthescaleandscopeofmanyICTdiffusionstudies.
Whilewehaveusedjustonetrademagazineforthisillustration,thecapabilityofthisapproachisnotlimitedinthenumberortypeofdiscourseoutlets.
Further,althoughourownknowledgehelpedvalidatethemethodsintheillustrationstudy,themethodsthemselvesdonotrelyonexpertknowledge.
Thisfeaturedifferentiatesourapproachfromotherclassificationmethodsbasedonexpertratingsoropinions(e.
g.
,[6],[16]).
Expertknowledgecanbeusefulforspecificresearchobjectives,butmethodsrelyingonexpertsarenotscalable.
Moreover,unlikescalableanalysisthatreliesonrelativelythinobservations,suchascitations(e.
g.
,[4])orvocabulary(e.
g.
,[1]),theKLdivergencemeasurecapturesboththevocabularyandtherichcontextoftheuseofvocabulary.
Overall,theseadvantagescreateamiddlegroundwherebothbreadthanddepthcanbebalancedinusefulways.
5.
3.
ImplicationsforICTdiffusionresearchThescalablecomputationalapproachwehavedemonstratedinthisstudycanhelptoexpandICTdiffusionresearchalongthreedimensions:innovations,datasources,andtime.
First,ourapproachcanhelpICTdiffusionresearchersovercomethelimitationofsingle-innovationdesignsbyconcurrentlyexploringmultipleinnovations.
Byfacilitatingthestudyofmultipleinnovationsandtheirrelationshipsoveranextendedperiodoftime,ourapproachenablesresearcherstodeveloptheoriesofICTdiffusiononamorerealisticfooting.
Therealityoftoday'sdigitaleconomyandinformationsocietycallsforanunderstandingofhowthediffusionprocessesofvariousICTsinteractwitheachother.
Inthisstudywehavefoundthatthepopularitytrajectoriesofsemanticallysimilarinnovationssometimesdemonstratesimilarpatterns(e.
g.
,Figures2and4)andsometimessuggestcompetitionorsubstitution(e.
g.
,Figures3and7).
Futureresearchshouldbuilduponthesefindingstofullyexplorethecontingencies.
Second,althoughweusedonlyonedatasource(InformationWeek)toillustratehowtheanalysisisdone,ourapproachisnotlimitedtothisoranyothersource.
Anysourcemaybebiasedbyitsownspecificsandthusitwouldbeusefultocollectdatafrommultiplesourcesandapplyourmethodologyinthesamewayaswedescribedabove.
Ontheonehand,astudymaydiscoveranobjectiverepresentationoftheinnovationrelationshipsbypoolingthedatafrommultiplesourcesinproportionsthatmayrepresentkeyconstituenciesofinnovationcommunities.
Ontheotherhand,researchersmayanalyzethedatacollectedfromeachsourceandcomparetheresults,revealingsimilaritiesanddifferencesamongvarioussegmentsoftheinnovationcommunitiesthatdifferentsourcesrepresent.
Forexample,apreviousstudyusingtheKLdivergenceandMDSmethodsfoundthattheICTinnovationrelationshipsdiscoveredinInformationWeekarticlesareverysimilartothosefoundinComputerWorldarticles[18].
Presently,weareintheprocessofcollectingandanalyzingdatafromothertrademagazinesandothersourcessuchasthepopularpress,newspapers,academicjournals,andinformalmedia.
Infact,weexpecttheclusteringstructurestodifferacrossdifferentsourcesduetotheirdifferentstylesorfoci(suchastheemphasesondifferentactorsintheinnovationcommunities).
ComparingtheresultsacrosssourceswillenrichourunderstandingoftheICTinnovationsaswellasthecommunitiesunderlyingtheseinnovations.
Third,duetothedynamicnatureofinnovationrelationships,itwouldbeusefultoconductaseriesof5Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010KLdivergenceandclusteringanalysesatmultipletimes.
TheevolvinghierarchicalstructureofICTinnovationswouldrevealwhatexactlyisdiffusing,asinnovationswithchangingmeaningsmightclusteratdifferenttimeswithdifferentinnovations.
ThisresearchcapabilitycouldallowdiffusionresearcherstoeventuallyunderstandthespeciationofICTinnovations[2],thatis,howvariouscategoriesofICTemerge,converge,anddivergeovertime.
Forexample,thetwoseparateclustersofweb2.
0technologiesthatwehavefoundinthisstudymayindicateaspeciationmoment,beyondwhichthetwotypesofinnovationscoulddiffuseindifferentways.
Tofurtherillustratetheutilityofourmethodologyinmulti-periodanalysis,weslicedourdatasetbyyearandperformedthesameanalysisoneachyear'sdata.
Thepagelimitdoesnotallowustoshowallclusteringresults.
Nevertheless,Figure9showsportionsoftheresultsforthreeyears.
Overall,thehierarchicalstructureofinnovationsdidchangeovertime.
Specifically,CRMmovedfromarelativelyseparateinnovationin1998toonecloselyassociatedwithERPin2001and2004.
However,incontrast,e-businessande-commerceareclusteredtogetherthroughouttheyears.
Insum,thisscalablecomputationalmethodologyenablesmulti-innovation,multi-source,multi-periodstudies,potentiallyadvancingknowledgeaboutthedynamicrelationshipsamongICTinnovations.
5.
4.
ImplicationsforICTadoptionanduseThisstudyprovidesthepractitionersattheforefrontofICTadoptionandusewithasetofscalabletoolsformonitoringandunderstandingnewandexistingICTinnovations.
Intheabsenceofexpertknowledge,asinthecasewherepractitionersconsideradoptinganewandpromisingICT,thesameanalysiscanbeappliedtothediscourseaboutthenewtechnologyandtothediscourseaboutexistinginnovationsaswell.
Thenewtechnology'sclustermembershipinthedendrogrammaythenservetosuggestitsbroadtype.
InusingICTs,practitionersmightalsofinditusefultoapplytheKLdivergenceandclusteringanalysistothediscoursefrominternalsourcessuchasemails,memos,manuals,andmeetingminutes.
TheresultingdendrogrammightthenserveastaxonomyofICTs.
ForprovidersofnewITproductsandservices,suchataxonomythatemergesfromdatamaycomplementtheproductcategoriesdesignatedinatop-downdesignprocess.
Thedata-driventaxonomyofICTscouldalsohelpidentifyredundanciesand/orestablishsharedunderstandingoftheICTsinuseacrossorganizationalunits.
6.
ConclusionInthisstudy,wehavedemonstratedascalablecomputationalapproach,basedonKLdivergenceandhierarchicalclustering,fordescribingandanalyzingtherelationshipsamongmultipleICTinnovations.
Inessence,weutilizesocialartifacts(i.
e.
,ICTdiscourse)tochartthetechnologicalterrainswhereICTsarediffusedandused.
Thissocio-technicalapproachwillrealizeitsfullpotentialincontinuedandsustainedresearchonthesocio-technicaldynamicsofICTdiffusion,adoption,anduse.
7.
AcknowledgementsThispaperisbaseduponworksupportedbytheNationalScienceFoundationunderGrantsNo.
IIS-0729459andSBE-0915645.
WewouldliketothankLidanWangforhersuggestiontousesymmetrizedKLdivergence.
8.
References[1]E.
AbrahamsonandM.
Eisenman,"Employee-ManagementTechniques:TransientFadsorTrendingFashions,"AdministrativeScienceQuarterly,vol.
53,no.
4,2008,pp.
719-744.
[2]R.
AdnerandD.
A.
Levinthal,"TheEmergenceofEmergingTechnologies,"CaliforniaManagementReview,vol.
45,no.
1,2002,pp.
50-66.
[3]J.
A.
BergerandC.
Heath,"IdeaHabitats:HowthePrevalenceofEnvironmentalCuesInfluencestheSuccessofIdeas,"CognitiveScience,vol.
29,no.
2,2005,pp.
195-221.
[4]L.
M.
A.
Bettencourta,A.
Cintron-Arias,D.
I.
Kaiser,andCastillo-Chavez,"ThePowerofaGoodIdea:QuantitativeModelingoftheSpreadofIdeasfromEpidemiologicalModels,"PhysicaA,vol.
364,no.
2006,2006,pp.
513-536.
[5]T.
M.
CoverandJ.
A.
Thomas,ElementsofInformationTheory,Wiley,NewYork,1991.
[6]P.
Ein-DorandE.
Segev,"AClassificationofInformationSystems:AnalysisandInterpretation,"InformationSystemsResearch,vol.
4,no.
2,1993,pp.
166-204.
[7]B.
Everitt,S.
Landau,andM.
Leese,ClusterAnalysis,4thed.
,OxfordUniversityPress,NewYork,2001.
[8]R.
G.
Fichman,"GoingBeyondtheDominantParadigmforInformationTechnologyInnovationResearch:EmergingConceptsandMethods,"JournaloftheAssociationforInformationSystems,vol.
5,no.
8,2004,pp.
314-355.
6Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010[9]S.
T.
FiskeandS.
E.
Taylor,SocialCognition,McGraw-Hill,NewYork,1991.
[10]P.
HansenandB.
Jaumard,"ClusterAnalysisandMathematicalProgramming,"Mathematicalprogramming.
,vol.
79,no.
1-3,1997,pp.
191-215.
[11]S.
KullbackandR.
A.
Leibler,"OnInformationandSufficiency,"TheAnnalsofMathematicalStatistics,vol.
22,no.
1,1951,pp.
79-86.
[12]C.
D.
ManningandH.
SchuTze,FoundationsofStatisticalNaturalLanguageProcessing,MITPress,Cambridge,MA,1999.
[13]D.
W.
Oard,"AWhirlwindTourofAutomatedLanguageProcessingfortheHumanitiesandSocialSciences,"inWorkingTogetherorApart:PromotingtheNextGenerationofDigitalScholarship,WashingtonDC:CouncilonLibraryandInformationResources,2009,pp.
34-42.
[14]N.
PhillipsandC.
Hardy,DiscourseAnalysis:InvestigatingProcessesofSocialConstructionvol.
50,SagePublications,ThousandOaksCA,2002.
[15]E.
M.
Rogers,DiffusionofInnovations,5thed.
,FreePress,NewYork,NY,2003.
[16]E.
B.
SwansonandN.
C.
Ramiller,"InformationSystemsResearchThematics:SubmissionstoaNewJournal,1987-1992,"InformationSystemsResearch,vol.
4,no.
4,1993,pp.
299-330.
[17]Y.
Takane,F.
W.
Young,andJ.
DeLeeuw,"NonmetricIndividualDifferencesMultidimensionalScaling:AnAlternatingLeastSquaresMethodwithOptimalScalingFeatures,"Psychometrika,vol.
42,no.
1,1977,pp.
7-67.
[18]C.
-J.
Tsui,P.
Wang,K.
R.
Fleischmann,D.
W.
Oard,andA.
Sayeed,"UnderstandingItInnovationsthroughComputationalAnalysisofDiscourse,"in30thInternationalConferenceonInformationSystems,Phoenix,AZ,2009.
[19]P.
Wang,"PopularConceptsBeyondOrganizations:ExploringNewDimensionsofInformationTechnologyInnovations,"JournaloftheAssociationforInformationSystems,vol.
10,no.
2,2009,pp.
1-30.
[20]P.
WangandN.
C.
Ramiller,"CommunityLearninginInformationTechnologyInnovation,"MISQuarterly,vol.
33,no.
4,2009,[21]R.
H.
WhittakerandS.
A.
Levin,Niche:TheoryandApplication,Downden,Hutchinson&Ross,Inc.
,Stroudsburg,PA,1975.
Table1.
ListofinformationandcommunicationtechnologyinnovationsAIArtificialIntelligenceMP3MP3playerASPApplicationServiceProviderMySpaceMySpaceATMAutomatedTellerMachineOLAPOnlineAnalyticalProcessingBIBusinessIntelligenceOSSOpenSourceSoftwareBlogBlogOutsourceOutsourcingBluetoothBluetoothPDAPersonalDigitalAssistantCADComputerAidedDesignRFIDRadioFrequencyIdentificationCRMCustomerRelationshipManagementSmartCardSmartCardDigiCamDigitalCameraSCMSupplyChainManagementDLearnDistanceLearningSFASalesForceAutomationDSLDigitalSubscriberLineSocNetSocialNetworkingDWDataWarehouseSOAService-OrientedArchitectureeBizeBusinessTelecommuteTelecommutingeComeCommerceTabletPCTabletPCEDIElectronicDataInterchangeUtiCompUtilityComputingERPEnterpriseResourcePlanningVirtualizationVirtualizationGPSGlobalPositioningSystemVPNVirtualPrivateNetworkGrpwareGroupwareWeb2.
0Web2.
0IMInstantMessagingWebServWebServicesiPhoneiPhoneWiFiWi-FiiPodiPodWikiWikiKMKnowledgeManagementWikipediaWikipediaLinuxLinuxYouTubeYouTubeMultimediaMultimedia7Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010Figure1.
DendrogramproducedbyhierarchicalclusteringanalysisoftheInformationWeekdataCluster1Cluster2Cluster3Cluster4Cluster58Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-20100500100015002000250030001998199920002001200220032004200520062007NumberofParagraphseBizeCom01002003004005006007008001998199920002001200220032004200520062007NumberofParagraphsSOAWebServFigure2.
Popularityofe-businessande-commerceFigure3.
PopularityofSOAandwebservices0501001502002503003504004505001998199920002001200220032004200520062007NumberofParagraphsVPNDSL0204060801001201401601998199920002001200220032004200520062007NumberofParagraphsSocNetMySpaceYouTubeFigure4.
PopularityofDSLandVPNFigure5.
Popularityofsocialnetworkinginnovations0501001502002501998199920002001200220032004200520062007Numberofparagraphsaboutweb2.
0,wiki,andwikipedia0100200300400500600700800NumberofparagraphsaboutblogWeb2.
0WikiWikipediaBlog0501001502002503003501998199920002001200220032004200520062007NumberofParagraphsiPhoneiPodFigure6.
Popularityofweb2.
0innovationswithusergeneratedcontentsFigure7.
PopularityofiPhoneandiPod9Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010-1.
50-1.
00-0.
500.
000.
501.
001.
502.
002.
501.
501.
000.
500.
00-0.
50-1.
00-1.
50-2.
00OLAPUtiCompSCMCADEDIVirtualizationSFADWGrpwareAIBISOAKMERPASPCRMeBizWebServOutsourceeComLinuxDLearnATMRFIDVPNDSLSmartCardTelecommuteTabletPCGPSPDAWiFiMultimediaiPodiPhoneDigiCamMP3Web2.
0BlogSocNetWikiWikipediOSSIMMySpaceYouTubeBluetooth-1.
50-1.
00-0.
500.
000.
501.
001.
502.
002.
501.
501.
000.
500.
00-0.
50-1.
00-1.
50-2.
00OLAPUtiCompSCM-1.
50-1.
00-0.
500.
000.
501.
001.
502.
002.
501.
501.
000.
500.
00-0.
50-1.
00-1.
50-2.
00OLAPUtiCompSCMCADEDIVirtualizationSFADWGrpwareAIBISOAKMERPASPCRMeBizWebServOutsourceeComLinuxDLearnATMRFIDVPNDSLSmartCardTelecommuteTabletPCGPSPDAWiFiMultimediaiPodiPhoneDigiCamMP3Web2.
0BlogSocNetWikiWikipediOSSIMMySpaceYouTubeBluetoothCADEDIVirtualizationSFADWGrpwareAIBISOAKMERPASPCRMeBizWebServOutsourceeComLinuxDLearnATMRFIDVPNDSLSmartCardTelecommuteTabletPCGPSPDAWiFiMultimediaiPodiPhoneDigiCamMP3Web2.
0BlogSocNetWikiWikipediOSSIMMySpaceYouTubeBluetoothFigure8.
MDSplotofthe47ICTinnovationsfrom10-yearInformationWeekdata199820012004Figure9.
PortionsofyearlydendrogramsproducedbyhierarchicalclusteringanalysisoftheInformationWeekdata10Proceedingsofthe43rdHawaiiInternationalConferenceonSystemSciences-2010
我们很多老用户对于BuyVM商家还是相当熟悉的,也有翻看BuyVM相关的文章可以追溯到2014年的时候有介绍过,不过那时候介绍这个商家并不是很多,主要是因为这个商家很是刁钻。比如我们注册账户的信息是否完整,以及我们使用是否规范,甚至有其他各种问题导致我们是不能购买他们家机器的。以前你嚣张是很多人没有办法购买到其他商家的机器,那时候其他商家的机器不多。而如今,我们可选的商家比较多,你再也嚣张不起来。...
imidc怎么样?imidc彩虹数据或彩虹网络现在促销旗下日本多IP站群独立服务器,原价159美元的机器现在只需要88美元,而且给13个独立IPv4,30Mbps直连带宽,不限制月流量!IMIDC又名为彩虹数据,rainbow cloud,香港本土运营商,全线产品都是商家自营的,自有IP网络资源等,提供的产品包括VPS主机、独立服务器、站群独立服务器等,数据中心区域包括香港、日本、台湾、美国和南非...
今天遇到一个网友,他在一个服务器中搭建有十几个网站,但是他之前都是采集站点数据很大,但是现在他删除数据之后希望设置可能有索引的文章给予404跳转页面。虽然他程序有默认的404页面,但是达不到他引流的目的,他希望设置统一的404页面。实际上设置还是很简单的,我们找到他是Nginx还是Apache,直接在引擎配置文件中设置即可。这里有看到他采用的是宝塔面板,直接在他的Nginx中设置。这里我们找到当前...
iphonewifi为你推荐
methoxychrome有人在认真做事路由route贵州省127支持ipad三星iphone空调iphone支持ipad支持ipadcolumnios5
虚拟主机评测网 山东虚拟主机 秒解服务器 idc测评网 贵州电信宽带测速 免费个人博客 圣诞节促销 嘉洲服务器 怎样建立邮箱 网站cdn加速 香港新世界中心 爱奇艺会员免费试用 最漂亮的qq空间 cloudlink 韩国代理ip 贵阳电信测速 摩尔庄园注册 服务器硬件配置 碳云 九零网络 更多