1
OnMaintainingSemanticNetworks:
Challenges,Algorithms,UseCases
Klaus Ulmschneider1, Bernd Michelberger2, Birte Glimm1,
Bela Mutschler2, Manfred Reichert1
1Ulm University, Germany
2University of Applied Sciences Ravensburg-Weingarten, Germany
Purpose–Knowledgeworkersareconfrontedwithamassiveloadofdatafromheterogeneous
sources,makingitdifficultforthemtodiscoverinformationrelevantinthecontextoftheirdaily
tasks.Asaparticularchallenge,enterpriseinformationneedstobealignedwithbusiness
processes.Inpreviouswork,theauthorsintroducedtheSemanticNetwork(SN)approachfor
bridgingthisgap,i.e.,fordiscoveringexplicitrelationsbetweenenterpriseinformationand
businessprocesses.Whathasbeenneglectedsofar,however,isSNmaintenance,whichis
requiredtokeepanSNconsistent,complete,andup‐to‐date.Thepapertacklesthisissueand
extendstheSNapproachwithmethodsandalgorithmsforenablingSNmaintenance.
Design/methodology/approach–ThepaperillustratesanapproachforSNmaintenance.
Specifically,theauthorsshowhowanSNevolvesovertime,classifypropertiesofobjectsand
relationscapturedinanSN,andshowhowthesepropertiescanbemaintained.Anempirical
evaluation,whichisbasedonsyntheticandreal‐worlddata,investigatestheperformance,
scalabilityandpracticabilityoftheproposedalgorithms.
Findings–Theauthorsprovethefeasibilityoftheintroducedalgorithmsintermsofruntime
performancewithaproof‐of‐conceptimplementation.Further,areal‐worldcasefromthe
automotivedomainconfirmstheapplicabilityoftheSNmaintenanceapproach.
Originality/value–Asopposedtoexistingwork,thepresentedapproachallowsforthe
automatedandconsistentmaintenanceofSNs.Furthermore,theapplicabilityofthepresented
SNmaintenanceapproachisvalidatedinthecontextofareal‐worldscenarioaswellastwo
businesscases.
Keywords–Semanticnetwork,knowledgerepresentation,evolution,maintenance,algorithm,
knowledge‐intensivebusinessprocess,technologymonitoring,decisionsupport
2
1 Introduction
Knowledgeworkersanddecisionmakersareconfrontedwithacontinuouslyincreasingloadof
datatheyhavetocopewithduringdailywork.Examplesincludee‐mails,officefiles,checklists,
guidelines,factsheets,webpages,andbestpractices.Indailypractice,knowledgeworkersand
decisionmakersarenotonlyinterestedingettingaccesstothisdata,buttheyalsorequire
comprehensiveandaggregatedinformationwhenperformingspecifictasksinabusiness
processcontext(Michelbergeretal.,2012a).Handlingsuchinformationisbyfarmorecomplex
andtime‐consumingthanjuststoringdata.Forexample,thisinformationmightbeincomplete,
incorrect,unreliable,unstructured,oroutdated(Michelbergeretal.,2011a;Rowley,2007).
Aparticularchallengeistoalignenterpriseinformationwithbusinessprocessesandtoprovide
relevantinformationtoknowledgeworkersanddecisionmakers.Inpractice,enterprise
informationisnotonlystoredindistributedandheterogeneoussources,butalsomanaged
separatelyfrombusinessprocesses.Forexample,shareddrives,databases,enterpriseportals,
andenterpriseinformationsystemsareusedtostoretheinformation.Inturn,business
processesandtheirtasksaremanagedusingprocess‐awareinformationsystems(Reichertand
Weber,2012).
Insuchanenvironment,informationandbusinessprocessesareoftenlinkedmanuallyaswell
asstatically.Forexample,ithasbeenshownthatenterpriseportalsrathercontaincomplexand
staticcontent(e.g.,longlistsofdocuments,largeprocessmaps).Inturn,thisratherconfuses
users(Hippetal.,2014).Therefore,itischallengingforthelattertoidentifytherelations
betweenenterpriseinformationandbusinessprocesses,whichiscrucialwhenperforming
specificprocesstasks.
Inpreviouswork,weintroducedtheSemanticNetwork(SN)approachbridgingthegap
betweenenterpriseinformationandbusinessprocesses(Michelbergeretal.,2013).SuchanSN
canbecreatedusinginabottom‐upmanner,i.e.,startingwiththeintegrationofenterprise
informationandbusinessprocessesfromheterogeneoussources(Michelbergeretal.,2012b).
Followingthis,theintegratedinformationandbusinessprocessesaresyntacticallyand
semanticallyanalyzed.TheresultingSNthencomprisesunifiedinformationobjects(e.g.,
checklists,guidelines,forms),processobjects(e.g.,pools,lanes,tasks,gateways,events),and
semanticrelations(e.g.,“issimilarto”,“isusedby”).Morespecifically,informationobjectsare
neededwhenperformingthetaskofabusinessprocess.Inturn,processobjectscorrespondto
processelements,suchastasksorgateways,thatguideprocess‐orientedwork.Finally,
semanticrelationsallowidentifyinginter‐linkedobjectsindifferentways,e.g.,information
objectsreferringtothesametopicorobjectsrequiredforperformingaparticularprocesstask.
AnSNconstitutesthebasisfortheprocess‐centricdeliveryofrelevantenterpriseinformation
toknowledgeworkersanddecisionmakers(cf.Fig.1).Morespecifically,anSNoffersan
applicationinterfacethatmaybequeriedtoretrievetherequiredinformation(Hippetal.,
2013).Basedonaquery(e.g.,“informationobjectsrelevantforcreatingareview”),theSN
automaticallydeliversrespectiveinformationobjectstousers(Michelbergeretal.,2012b).
3
Figure1:Deliveringrelevantinformationobjects
Inordertoprovidethatprocessinformationtoprocessparticipantsfittingbesttotheircurrent
demands,theinformationandprocessobjectscapturedinanSNneedtobecomplete,
consistent,andup‐to‐date.Consequently,anSNmustbecontinuouslymaintained.Intwocase
studiesaswellasanonlinesurvey(Hippetal.,2011;Michelbergeretal.,2011b),wehave
alreadyshownthatSNmaintenanceisaprerequisitetobeabletocontinuouslyprovidethe
requiredinformationtoknowledgeworkersanddecisionmakers.SNmaintenance,however,is
anon‐trivialtask.Forexample,objectsmaybeadded(e.g.,newguidelinesarecreated),
updated(e.g.,aprocesstaskismodified),ordeleted(e.g.,achecklistisnolongervalid).
Likewise,relationsmaybeestablished(e.g.,whendiscoveringthattwodocumentshavethe
sameauthororarestoredinthesamefileformat),updated(e.g.,twoformsbecomemore
similartoeachother),ordeleted(e.g.,twodocumentshavenolongerthesameauthor).On
onehand,suchchangesmayhappenoutsidetheSN(e.g.,achecklistmayhavebeenchangedin
adatabase),i.e.,thechangeisexogenous.Ontheother,changesmayoccurinsidetheSN(e.g.,
alifecyclestatuschangeofanobjectlikeaguidelinebecomingoutdated).Thesechanges,in
turn,aredenotedasendogenous.Bothexogenousandendogenouschangesmustbeproperly
handledbytheSN(cf.Fig.2).
Figure2:Exogenousandendogenouschanges
PickinguptheaforementionedchallengesregardingSNmaintenance,thecontributionofthis
paperisasfollows:First,weproposeanapproachforSNmaintenance.Specifically,weshow
howSNsevolveovertimeandidentifycharacteristicsofobjectandrelationpropertiesaswell
astheirinfluencewithrespecttoSNmaintenance.Second,weintroducethreealgorithms
dealingwithexogenousandendogenouschangesofanSN.Inthiscontext,weexaminethe
feasibilityandcostsofthealgorithmsthroughaproof‐of‐conceptimplementation.Further,we
demonstratebasedonanempiricalevaluationintheautomotivedomain,thatautomatedSN
maintenanceisessential,whilebeingpracticalatthesametime.Notethatthispaperprovides
anextendedversionoftheworkwepresentedinMichelbergeretal.(2014).Theadditional
contributionsareasfollows:
Thepaperillustratestheuseofthethreealgorithmsalongtworeal‐worldusecases.The
4
firstoneillustratesthemonitoringoftechnologiesintheautomotivedomain,whilethe
secondonedealswithdecisionmakinginthecontextoftechnologymanagement.
Thepapercomprisesadetaileddescriptionofthesixphasestobepassedwhencreating
anSN.Inparticular,thesixthphasecorrespondstoSNmaintenance.
ThepaperprovidesadetaileddiscussionontheadvantagesanddisadvantagesoftheSN
approach.Further,itprovidesasubstantiallyextendeddiscussionofrelatedwork.
Thepaperprovidesdeeperinsightintothecasestudy.
Theremainderofthepaperisorganizedasfollows.Section2introducesthepreliminaries.
Section3thenpresentstheSNmaintenanceapproach.Section4validatesthecostsofthe
algorithmsandpresentsacasestudydemonstratingtheirapplicability.Section5dealswiththe
applicationofthemaintenanceapproachbasedontworeal‐worldusecases.Finally,Section6
discussesrelatedworkandSection7concludesthepaper.
2 Preliminaries
AnSNconstitutesalabeledandweighteddirectedgraphwhoseverticesrepresentobjectsand
the(labeled)edgesrepresentthesemanticrelationsbetweentheobjectswithweights
indicatingtherelevanceoftherelations.Aweightisexpressedintermsofanumberranging
between0and1,with1indicatingthestrongestpossiblesemanticrelation.
Definition1(SemanticNetwork):ASemanticNetworkSNisatuple,,,( LEV ),, wl ffW ,where
VisasetofverticessuchthateachVvrepresentsaninformationobjectoraprocessobject;
E
isamultisetofedgessuchthateachedgeEvve
),(= ,Vvv
,andvv
,representsa
relationbetweensuchobjects.ThefunctionLEfl:labelseachedgeEewithanedge
labelfromthesetoflabels
L
.Furthermore,thefunctionWEfw:assignsaweightfromthe
setofweightsWtoeachedgeEe.GivenanedgeEvve
),(= ,wecall
v
thesourceandv
thedestinationofe.
Definition2(Neighborhood):GivenavertexVv,theinternalneighborhoodof
v
,denoted
)(v
,isthesetofvertices}),(|{ Evvv
.Analogously,forVv,theexternalneighborhood
ofv,denoted)(v
,isthesetofvertices}),(|{ Evvv
.Then,thetotalneighborhoodof
Vvistheunionoftheinternalandexternalneighborhoodof
v
,denoted)(v.
Definition3(Degree):TheincomingdegreeofavertexVvisthenumberofincomingedges
andtheoutgoingdegreeisthenumberofitsoutgoingedges.Thetotaldegreeof
v
isthesum
ofitsincomingandoutgoingdegree.
Forexample,giventwoedgesEvvevve
),(=),,(= ,Vvvv
,, ,wecallvaninternal
neighborofvandv anexternalneighborofv.Thus,thetotaldegreeofvis2.
Notethatweoftenrefertoverticesasobjects(e.g.,informationandprocessobjects)andto
edgesasrelationsofanSN.Wenextdefinepropertiesforverticesandedges.
5
Definition4(Properties):EachvertexVvandeachedgeEehasasetofproperties)(vP
and)(eP ,respectively,whereeach)()( ePvPp isapair),( valkey .Wedenotekeyasthe
uniquenameandvalasthevalueofpandwrite)(vkey ()(ekey )todenoteval .
InordertocreateanSN,businessprocessesandpiecesofinformation,possiblyfromdifferent
datasources(e.g.,processrepositories,shareddrives),aretransformedintoprocessand
informationobjects(cf.Figs.3(a)and3(b)),eachrepresentedbyavertexanditsaccording
properties.Thetransformationensuresthatproprietaryformats(e.g.,officeformats)are
convertedintoauniformformat,whichallowsanalyzingtheSNobjects.
Afterthat,SNobjectsaresyntacticallyandsemanticallyanalyzedtodetecttheirsemantic
relations(cf.Fig.3(c))(Hippetal.,2013).First,properties(e.g.,authorship)arecompared
(syntacticanalysis),e.g.,tolinkobjectswiththesameauthor.Second,thepropertiesofthe
objectsareanalyzed(semanticanalysis).Forthispurpose,algorithmsfromthefieldsofdata
mining,textmining(e.g.,textpreprocessing,linguisticpreprocessing,clustering,classification,
informationextraction),pattern‐matching,andmachinelearning(e.g.,supervisedlearning,
unsupervisedlearning,reinforcementlearning,transduction)areapplied(Hothoetal.,2005;
Wurzer,2008)inordertofurtherclassifyandgroupcorrelatedobjects.
Figure3:SchematiccreationofanSN
SemanticrelationsinanSNexistbetweeninformationobjects(e.g.,aguidelinesimilarto
anotherone)orprocessobjects(e.g.,aneventtriggeringasub‐process).Additionally,semantic
relationsexistbetweeninformationandprocessobjects(e.g.,aninstructionrequiredfor
executingaspecificprocesstask).
Generally,anSNiscreatedinsixconsecutivephases(cf.Fig.4)followingabottom‐up
approach,i.e.,westartwiththeintegrationofbusinessprocessesandenterpriseinformation
thatoriginatingfromheterogeneoussourcessuchasdatabases,shareddrives,enterprise
portals,processrepositories,orenterpriseinformationsystems(Michelbergeretal.,2013).
InPhase1,businessprocessesrelevantforanSNneedtobeidentifiedandintegrated.Inthis
context,relevancydependsonwhichprocessesshallbesupportedbytheSN.Thebusiness
processestobeintegratedmustbeexplicitlyspecified,e.g.,usingaprocessmodelinglanguage
suchasBPMN(FreundandRücker,2012)orEPC(Scheer,2002).Onlysuchanexplicitprocess
descriptionallowsfortheautomatedtransformationofaprocessschemaanditscorresponding
processinstancesintoprocessobjects.Forthispurpose,allrelevantprocessobjects(e.g.,tasks,
events,gateways,dataobjects,pools,lanes,sequenceflows,messageflows,orassociations)
areidentified.Inturn,theresultingobjectsarethenusedtocreatetheSN'sfirststageof
6
expansion.InPhase2,relevantprocessinformation(e.g.,e‐mails,officefiles,manuals,
templates,forms,checklists,orguidelines)isaddedtotheSN;i.e.,thealreadyexistingSNis
extendedbyaddinginformationobjectsofdifferentgranularitylevels,rangingfromfine‐grained
information(e.g.,databasetuple)tocoarse‐grainedone(e.g.,multi‐pageofficedocument).
Figure4:ConstructionphasesoftheSN
InPhase3,therelationsamongtheprocessobjectsareidentified,i.e.,processobjectssuchas
sequenceflows,associationsormessageflowsaretransformedintoprocessobjectrelations.In
Phase4,theinformationobjectrelationsbetweentheSNinformationobjectsarediscovered.
Explicitrelations,likehyperlinksindocuments,arediscoveredfirst.Then,algorithmsfromthe
fieldsofdatamining,textmining,pattern‐matching,andmachinelearningareappliedto
discoverimplicitrelationsaswell(Hothoetal.,2005;Wurzer,2008).InPhase5,inturn,cross‐
objectrelationsbetweeninformationandprocessobjectsareidentified.Forthispurposesimilar
algorithmsasusedinPhases3and4areapplied.Inaddition,pre‐definedbusinessrules(e.g.,
conditionalconstraints,derivationsorprocessrules)areusedtodetectfurtherrelations
(Michelbergeretal.,2012b;Wurzer,2008).Finally,inPhase6,theSNismaintained.Thisphase
dealswiththecontinuousintegrationaswellasthecontinuousanalysisofinformationand
processobjectsincludingtheircorrespondingrelations.NotethatSNmaintenanceconstitutesa
prerequisiteforprovidingrelevantandup‐to‐dateinformationtoknowledgeworkersand
decisionmakers.Inthefollowing,wepresentconceptsandalgorithmsusedinthemaintenance
phase.
3 MaintainingSemanticNetworks
ThissectionintroducestheSNmaintenanceapproach.First,weshowhowanSNevolvesover
time(cf.Section3.1).Second,weillustratewhypropertiesofobjectsandrelationsare
important(cf.Section3.2).Third,wepresentacollectionofalgorithmsforproperlymaintaining
SNs(cf.Section3.3).
3.1 Evolution
SNevolutionisdrivenbyexogenousaswellasendogenouschanges.These,inturn,resultin
changesofobjectsandrelationsincludingtheirpropertyvalues.Therefore,wefurther
distinguishbetweenevolutionindepthandbreadth(OerteltandUlmschneider,2013).Depthis
7
definedbythesizeofallpropertyvaluesofanSN,i.e.,theamountofinformation(e.g.,the
informationstoredwithrespecttoallSNobjects).Breadth,inturn,isdefinedasthenumberof
relationshipsinanSN,i.e.,thecardinalityofthesetofedges.
Depthmaybeincreasedbyaddingobjects(e.g.,newdocumentsonashareddrive),adding
properties(e.g.,addingkeywordstoaobject),orupdatingpropertyvalues(e.g.,describinga
propertyinmoredetail)(cf.Fig.5(a)).Inturn,deletingobjectsandpropertiesdecreasesthe
depthofanSN(cf.Fig.5(b)).Notethatupdatesofpropertyvaluesmightdecreasedepthaswell.
Breadthcanbeincreasedbyaddingrelations,e.g.,addingalinkbetweentwoobjects(cf.Fig.
5(c)).Bycontrast,deletingrelations(e.g.,twoobjectsnolongerhavethesameauthor)
decreasesbreadth(cf.Fig.5(d)).Hence,depthandbreadthareindicatorsforthecostof
performingmaintenancetasks.
Figure5:SNevolutionindepthandbreadth
InordertoformalizeSNmaintenanceoperations,weneedacomponentwhichiscapableto
adaptexogenousandendogenouschangestotheSN.Weachievesuchfunctionalitybythe
conceptofanaction.
Definition5(Action):AnactionchangesanSNinastructuredandstandardizedway.Each
actionahasasetofparameters)(aPA ,whereeach)(aPApaisapair),( valkey .Wecall
keytheuniquenameandvalthevalueofpa andwrite)(akey todenoteval .Aparameter
pa ,inturn,iseithermandatoryoroptional.Ifpa withkeykeyismandatory,then,foreach
actiona,thereexistsavaluea
val suchthat)(),( aPAvalkey a.
Typicalmandatoryparametersofanactionareauniqueidentifieruri(e.g.,aURL)andthe
functionfunc(e.g.,add,update,delete)tobeexecuted.Actionsaretriggeredbyexogenousor
endogenouschanges(cf.Fig.6),e.g.,whenadocumentonashareddriveisdeleted(an
exogenouschange)orbecomesoutdated(anendogenouschange).Accordingly,respective
eventstriggeradd,updateanddeleteoperationsontheSN.Therefore,actionsadaptinternalas
wellasexternaleventsandaffecttheSN.
Figure6:ActionschanginganSN
Asexampleconsideranengineerintheautomotivedomainconductingareviewofproduct
8
requirementsdocumentedinfunctionalspecifications.Thegoalistoimproveaswellasto
approvesuchspecifications.Duetoarevisionofthereviewprocess,anemployeefromthe
qualitymanagementdepartmentreplacedanoutdatedreviewtemplate.Thus,anaction1
awas
triggeredwith=)( 1
auri “H:/templates/review‐v1.xls”and=)( 1
afunc “delete”.Thereby,
anotheraction2
awastriggeredwith=)( 2
auri “H:/templates/review‐v2.xls”and=)( 2
afunc
“add”.However,basedonnewguidelinestheengineernoticedthatthetemplatewas
incomplete(e.g.,arequiredquestionwasmissing).Therefore,headaptsit.Thus,anotheraction
3
aistriggeredwith=)( 3
auri “H:/templates/review‐v2.xls”and=)( 3
afunc “update”.
3.2 PropertyClassification
SNmaintenancenotonlyrequirestoconsidertheobject‐relation‐level,butthepropertiesof
objectsandrelationsaswell.Furthermore,iftheauthorofadocumentchanges,forexample,it
isnotnecessarytoupdatetheentireobjectbutonlyrelevantparts,i.e.,thevalueofproperty
authorandaccordingrelations.Onecanobservethatcertainpropertieschangeovertime(e.g.,
filesize),whereasothersdonot(e.g.,uri).Thiscanbeexploitedinmaintenancealgorithmsby
focusingonthosepropertiesbeingrelevantforaparticularoperation.Tocapturethis,we
categorizepropertiesasfollows:
Definition6(ExistenceandMutability):Propertiesareclassifiedintotwocategories:existence
andmutability.Existenceexpresseswhetherapropertyismandatoryoroptional,where
propertypwithkeykeyismandatoryforverticesV(edges
E
)ofanSNif,foreachVv(
Ee),thereexistsavaluev
val (e
val )suchthat)(),( vPvalkey v()(),( ePvalkey e),
otherwise,itisoptional.Mutability,inturn,expresseswhetheraproperty’svalueisdynamicor
static,wherepisdynamicifvalin),( valkey canchangeovertimeanditisstaticotherwise.
Forexample,forVv,typicalmandatorypropertiesareauniqueidentifieruri,adatasource
source,acreationdatecdate,amodificationdatemdate,andacontentcont.Thecategories,
existenceandmutability,canbecombinedintoamatrixcomprisingfourblockstowhichwe
assigntheproperties(cf.Fig.7).
Figure7:Propertyclassification
9
Inthefollowing,weillustratetheassignmentofindividualpropertiestodifferentblockswith
examples:
(a) mandatory/dynamic:SomepropertiesarealwayspartofanSNandaredynamic.For
example,themodificationdatemdatechangeswitheveryupdateofanobjectorrelation.
Thecontentcontorthetotaldegreedegofanobjectcanchangeovertimeaswell.
Therefore,thesepropertiesaremandatoryanddynamic.
(b) optional/dynamic:Thetitleofadocumentcanchangeovertime,butsomefiletypes(e.g.,
atextfile)donothaveatitle.Therefore,thetitleofadocumentcanbeconsideredas
optionalanddynamic.Apropertycontainingprojectbudgetsinvestmightnotbe
availableforallverticesoredgesandcanvaryovertimeaswell.Therefore,investcanbe
consideredasoptional/dynamicaswell.
(c) mandatory/static:Anidentifieruri,acreatororacreationdatecdateexistsforallobjects
andrelationsandthereforeismandatory.Sincethesepropertiesdonotchangeover
time,theycanbeconsideredasstatic.
(d) optional/static:Ifapropertydoesnotchangeovertimeanddoesnotexistforevery
objectorrelationitiscalledoptional/static,e.g.,thefiletypeofadocument.
Basedonthepropertyclassificationweinferthefollowingforadding,updating,anddeleting
elementsofanSN:Onemustensurethatstatic/mandatoryproperties(c)aregivenasa
minimumrequirementwhenaddingelements(cf.Fig.8(a)).Whendeletingelementsonehasto
considerpropertieswithinallblocks(a,b,c,d)(cf.Fig.8(b)).Whenexecutingupdates,only
propertiesnotassignedtothemandatory/staticblock(a,b,d)mustbeconsidered(cf.Fig.8(c)).
NotethatthegreybackgroundcolorinFigure8indicatesaffectedblocksforeachfunction(cf.
Section3.3).
Figure8:Propertyclassificationandfunctions
Therefore,fromtheevolutionaryperspective(cf.Section3.1),dynamicpropertiescapture
changesofexistingverticesandedgesinanSN,whereasstaticpropertiescoverchangesin
termsofaddedverticesandedges.Additionally,mandatorypropertiesalwaysallowfor
analyses,likecomparisonsontheoverallsetofSNverticesandSNedges.Usually,dynamic
propertiesstoreprocessinformationwhereasstaticpropertiesusuallystoremetadata.
3.3 Algorithms
InordertosuccessfullymaintainSNs,wefirstspecifyfunctions(add,delete,andupdate)that
canbetriggeredbyactionstoperformmaintenanceoperationswiththealgorithms.
10
Theadd‐functionaddsavertexadd
vanditspropertiestoaSemanticNetworkSN and
determineswhichsemanticrelationsexistbetweenadd
vandexistingvertices.Asmentioned
above,mandatory/staticpropertiesaretheminimuminputparameterfortheadd‐function.
Require:),,,,,(= wl ffWLEVSN anSN,add
vthevertextobeaddedincl.itsproperties
)( add
vP ;
Ensure:SN isupdated;
}{:= add
vVV ;
foreachVvdo
if)()( add
vurivuri then
} and between edge/s new{:= add
vvEE ;
endif
endfor
Functionadd( add
vSN,)
Thedelete‐functiondeletesavertexdel
vinaSemanticNetworkSN includingexisting
semanticrelationsbetweendel
vanditstotalneighborhood.Notethatthefunctionconsidersall
blocksofthepropertyclassification,i.e.,allpropertiesofdel
varedeleted.
Require:),,,,,(= wl ffWLEVSN anSN,
del
vthevertextobedeletedincl.itsproperties
)(
del
vP
;
Ensure:SN isupdated;
:=
v
getVvs.t.)(=)( del
vurivuri ;
)}(|),(),,{(\:= vvvEE
;
}{\:= vVV ;
Functiondelete( del
vSN,)
Theupdate‐functiontakesavertexupd
vasinput,whichisusedtoupdatethevertexvinthe
SNthatisidentifiedbythesameuriasupd
v.Thefunctionalsoadds,deletes,andupdates
semanticrelationsbetweentheupdatedvertexvandexistingvertices.Notethat
mandatory/staticpropertiesarenotconsideredincaseofobjects.
11
Require:),,,,,(= wl ffWLEVSN anSN,upd
vthevertexusedtoupdatetheSNincl.its
properties)( upd
vP ;
Ensure:SN isupdated;
}staticmandatory/not is |)({:=)( pvPpvP updupd ;
:=
v
getVvs.t.)(=)( upd
vurivuri ;
foreach)()val,key( vPdo
if)()val,key( upd upd
vPthen
upd
valval := ;
)},{(\)(:=)( upd updupd valkeyvPvP ;
else
)},{(\)(:=)( valkeyvPvP ;
endif
endfor
)()(:=)( upd
vPvPvP ;
foreachVv
do
if)(vv
then
:=Eupdateedge/sbetw.vandv;
} and between edge/s obsolete{\:= vvEE ;
endif
if )()( vurivuri
then
} and between edge/s new{:= vvEE
;
endif
endfor
Functionupdate( upd
vSN,)
Basedonthesefunctions,weproposethreealgorithmsformaintainingSNs.Themaintenance
isbasedontwomainprinciples:thepush‐andthepull‐principle.Theformercanbeappliedto
bothexogenousandendogenouschanges,whereasthelattercanonlybeappliedtoexogenous
changes.
Withthepush‐principle,thedatasourcepushesinformationandbusinessprocesses
automaticallytoanSNwhentheyareadded,updated,ordeletedwithinthedatasource.
However,withregardtoexogenouschanges,prerequisiteisthatthedatasourceisabletosend
notificationsifinformationand/orbusinessprocesseshavebeenchanged.Regarding
endogenouschanges,theprerequisiteisthattheSNdetectschangesautomaticallyandtriggers
respectiveactions.
Withthepull‐principle,anSNgathersinformationandbusinessprocessesfromadatasource.
Suchamaintenanceprocessistriggeredbytime‐basedschedulers,i.e.,theSNismaintainedat
acertainpointintime.Theprincipleisusedfordatasourceswhicharenotcapableofsending
changenotifications(e.g.,adocumenthaschanged)totheSN.
Altogether,theuseofaspecificprincipledependsonthecapabilitiesofadatasource(whether
12
thedatasourceisabletosendchangenotificationstotheSNornot).Forexample,foran
enterpriseinformationsystemwhichiscapableofsendingnotificationsweusethepush‐
principlewhereasforashareddriveweusethepull‐principle.
Foreachofthesetwoprinciples,weintroduceacorrespondingalgorithm(cf.Fig.9).
PrerequisiteforbothprinciplesisthattheSNhasaccesstounderlyingdatasources.Incaseof
exogenouschangestheSNtransformsinformationandbusinessprocessesintoauniform
format.Incaseofendogenouschangesnotransformationisnecessary.
Figure9:Push‐andPull‐Algorithm
3.3.1 Push‐Algorithm
ThePush‐AlgorithmdealswithchangesofanSNbasedonthepush‐principle,e.g.,apolicyisno
longervalidin2015andthecorrespondingobjectintheSNhastobemaintainedaccordingly.
Thus,themaintenanceoftheSNistriggeredbyanactionthatisappliedtotheSNbythePush‐
Algorithm.
Thealgorithmworksasfollows:Intheaddandupdatecase,wecreateavertexvincludingits
propertiesfromthedatasourceaffectedbytheactiona(i.e.,basedontheurioftheaction).
Afterthat,wecallthecorrespondingfunction.Inthedeletecase,weidentifythecorresponding
vertexVvbasedontheurioftheactionandcalltheaccordingfunction.Hence,thePush‐
AlgorithmappliesendogenousandexogenouschangestotheSNbyactions.
13
Require:),,,,,(= wl ffWLEVSN anSN,aanaction;
Ensure:SN isupdated;
switchfunc(a)do
caseadd
:=
v
createavertexincl.itspropertiesfromthedatasourceaffectedby
theuri ofa;
),( vSNadd ;
break;
endcase
caseupdate
:=
v
createavertexincl.itspropertiesfromthedatasourceaffectedby
theuri ofa;
),( vSNupdate ;
break;
endcase
casedelete
:=
v
getVvs.t.)(=)( aurivuri ;
),( vSNdelete ;
endcase
endswitch
Algorithm1:Push‐Algorithm
3.3.2 Pull‐Algorithm
ThePull‐AlgorithmdealswithchangesofanSNbasedonthepull‐principle,i.e.,datahas
changedinthedatasourceandneedstobegatheredbytheSN.Forexample,documentsona
shareddriveareupdatedand,therefore,respectivechangeshavetobemadeintheSN.As
aforementioned,themaintenanceoftheSNistriggeredbyascheduler.Thealgorithmworksas
follows:First,wecreateasetofverticesds
Vfromadatasourceds.Afterthat,foreachvertex
VvintheSNthatwascreatedfromds(propertysource(v)),wecheckifacorresponding
vertexdsds Vv exists.Ifthisisthecase,wecheckifds
v
isnewerthan
v
(e.g.,bycomparingthe
creationand/ormodificationdates).If
v
isout‐of‐date,itisupdatedwiththepropertiesofds
v
bycalling),( ds
vSNupdate .Afterthat,ds
visremovedfromds
V.Ifnocorrespondingvertexexists
inthedatasource,wedeletethevertex(v)intheSNusingthe),( vSNdelete function.Finally,we
addeachremainingvertexdsds Vv fromthedatasourcetotheSNbycalling),( ds
vSNadd ,
whichleavestheSNsynchronizedwiththedatasource.Hence,thePull‐Algorithmallowsfor
maintainingtheSNatacertainpointintime.Notethattheprocesshastoberepeatedforeach
datasourcewhichmustbesynchronized.
14
Require:),,,,,(= wl ffWLEVSN anSN,dsthedatasource;
Ensure:SN isupdated;
:=
ds
Vcreateasetofverticesincl.theirpropertiesfromthedatasourceds;
foreachVvs.t.dsvsource =)( do
:=
ds
vgetdsds Vv s.t.)(=)( vurivuri ds ;
ifds
v
then
ifds
visnewerthanvthen
),( ds
vSNupdate ;
endif
}{\:= dsdsds vVV ;
else
),( vSNdelete ;
endfor
foreachdsds Vv
),( ds
vSNadd ;
endfor
Algorithm2:Pull‐Algorithm
3.3.3 Partial‐Pull‐PrincipleandAlgorithm
Inpractice,SNscancomprisealargeamountofobjectsandrelations.MaintainingSNsusing
thepull‐principlecan,therefore,beaverytime‐consumingtask.Inaspecificworkcontext,a
usermight,however,onlybeinterestedinaselectedpartoftheSN.Forexample,duringa
review,reviewtemplates,reviewminutes,existingreviews,orevenresultsofareal‐time
evaluation(e.g.,prioritizationofprojectsinaworkshop)areofgreatimportance,while
checklistsandbestpracticesforperformingeffectiveprojectmanagementarelessinteresting.
Thus,itissufficienttomaintainonlytheseobjectsandrelationsthatarerelevanttotheuser
whenqueryingtheSN.Tocapturethis,weintroduceafurtherprinciple,calledthepartial‐pull‐
principle,wheretheSNgathersonlyinformationandbusinessprocessesfromdatasourcesas
requestedbyauser.These(andonlytheseobjects)arethenupdatedondemand.
Incontrasttotheotherprinciples,thepartial‐pull‐principleiscompletelyuser‐drivenbecauseit
istriggeredbyauserrequest,whereasthepush‐ andpull‐principlearemachine‐driven,e.g.,
throughnotificationsfromotherenterpriseinformationsystemsorschedulers.
Basedonthepartial‐pull‐principle,weintroduceathirdalgorithm(cf.Fig.10)asalightweight
versionofthePull‐Algorithm.ItdoesnotmaintaintheentireSN,butonlythepartswhichare
relevantforagivenrequest.
15
Figure10:Partial‐Pull‐Algorithm
Thealgorithmworksasfollows:First,wecreateasetofverticesds
V
fromaffecteddatasources
accordingtotheuserrequestreq .Then,foreachvertexVvaffectedbyreqweretrievethe
correspondingvertexds
vfromtheaffecteddatasources.Thus,ifacorrespondingvertexds
vis
inthedatasource,wecheckifds
visnewerthanv(e.g.,bycomparingthecreationand/or
modificationdates)andifvisout‐of‐date,itisupdatedwithds
v
bycalling),( ds
vSNupdate .If
thereisnocorrespondingvertexinthedatasource,wedeletethevertex(v)intheSNusingthe
),( vSNdelete function.Hence,thePartial‐Pull‐AlgorithmallowsformaintainingpartsofanSN
basedonauserrequestandensuresthatallrequestedobjectsincludingtheirpropertiesare
synchronizedwithaffecteddatasources.
Require:),,,,,(= wl ffWLEVSN anSN,reqtherequesttoanSN;
Ensure:SN ispartiallyupdated;
req
Vcontainstherequestedvertices;
:=
ds
Vcreateasetofverticesfromthedatasourcesaffectedbyreq;
foreachVvaffectedbyreqdo
:=
ds
vgetdsds Vv s.t.)(=)( vurivuri ds ;
ifds
vthen
ifds
visnewerthanvthen
),( ds
vSNupdate ;
:=
upd
vgetVv
s.t.)(=)( ds
vurivuri ;
}{:= updreqreq vVV ;
else
}{:= vVV reqreq ;
endif
else
),( vSNdelete ;
endif
endfor
Algorithm3:Partial‐Pull‐Algorithm
16
4 Validation
ThissectionshowsthatthealgorithmsareabletosuccessfullymaintainanSN.Forthis
validation,weimplementedthealgorithmsandevaluatedtheirperformanceconsideringdepth
andbreadth,i.e.,wemeasuredthetimeneededtoadd,updateanddeleteobjectsaswellas
relations.Wefurtherevaluatedtherelevanceaswellastheapplicabilityofthealgorithmsinthe
contextofacasestudywhichweconductedintheautomotivedomain.
Thevalidationwasguidedbythreeresearchquestions(cf.Tab.1):
# Research Questions
RQ1 Is automatic maintenance of an SN feasible considering both exogenous and
endogenous changes?
RQ2 How do depth and breadth affect the runtime of the SN maintenance algorithms?
RQ3 How essential is automatic maintenance of SNs in practice?
Table1:ResearchQuestions
4.1 ImplementationandConfiguration
DrivenbytheresearchquestionspresentedinTable1wecreated(a)awell‐definedsetofSNs
forevaluatingtheperformancewithsyntheticdata(cf.Fig.12)toanswerRQ1andRQ2aswell
as(b)aspecificSNforempiricalevaluatingthealgorithmswithrealbusinessdatatoanswer
RQ3.Notethattheobjectsandrelationssharedidenticalpropertiesinordertoenable
comparabilityregardingmeasurementsaswellasenabletheevaluationoftheused
configurationinarealworldapplication.
Werealizedtheprototypeasa3‐tierarchitecture.Thepresentationlayerwasimplemented
withthewebapplicationframeworkPlay1,theTwitterBootstrapframework2,theData‐Driven
Documents(D3)library3,jQuery4,HTML5templates,andCascadingStyleSheets(CSS3).We
createdtheSNsusingthesemanticmiddlewareiQserGINServer(v.2.0.0.36)(Wurzer,2008).In
addition,wedevelopedJavaopen‐sourceplugins5onthelogiclayer.Thedatalayerwasbased
onaLucenesearchindex6andaMySQLdatabase7.
BasedonthepropertyclassificationdescribedinSection3.2,weconfiguredobjectsand
relationsoftheprototype.Asmandatory/staticobjectproperties,wechosecdate,uri,and
source(cf.Fig.11(a)).Optionalpropertieswerefiletypeandtitle:thefiletypedoesnotchange
overtime(static),whereasthetitlemayevolve(dynamic).Furthermore,thetitlemightnotbe
availableforeveryfiletype(e.g.,textfile),i.e.,itisoptional.Mandatorydynamicproperties
includedbuzz(i.e.,useractivityonobjects),cont,deg,andmdate.
17
Figure11:Propertyclassificationimplementation
Analogoustoobjectproperties,uriandcdateweremandatoryforrelations(cf.Fig.11(b)).
However,relationshadadditionalmandatorypropertiessuchassourcevertexanddestination
vertex.Thelabelofanedgewasalsomandatoryandstatic.However,theweightofanassigned
labelmayvaryovertimeand,therefore,itwasconfiguredasmandatoryanddynamic.As
example,changingcontmayaffecttheweightof“issimilarto”relations.Additionally,the
reasonofarelation,whichdescribeswhyarelationwasestablished(e.g.,aparticularmethod
thatdetectedan“issimilarto”relation),wasclassifiedasstaticandoptional.AllSNswere
createdwiththefollowingrelations:“isauthorof”,“hassametitleas”,and“issimilarto”.
Aftersettinguptheprototype,wefirstvalidatedtheperformanceofthealgorithms(cf.Section
4.2),i.e.,theinfluenceofdepthandbreadthonthealgorithms.Theperformancetestswere
executedonamachinewithanIntelquad‐coreCPUIntelCorei72670Qwith3.1GHz,16GB
RAM,512GBsolid‐statedrive(SATA6Gbit/s),andaWindows764‐bitoperatingsystem.Then
weevaluatedtheapplicationofthealgorithmsinthecontextofacasestudyintheautomotive
domain(cf.Section4.3).
4.2 TechnicalValidation
ThesuccessfulimplementationandinitialtestshavealreadydemonstratedthatautomaticSN
maintenanceisfeasibleingeneral.Wenowexaminetheruntimeinconsiderationofdepthand
breadth.InordertoaddressresearchquestionsRQ1andRQ2,weinvestigatedtheperformance
ofadd,updateanddeleteoperationsforthePull‐aswellasthePush‐Algorithm(cf.Section3.3).
Followingthis,wecreatedSNscomprising5,50and500objects,oncewithsmall(1KB)and
oncewithlarger(100KB)textfiles(cf.Fig.12).Toobtaincomparableresultsforthe
measurements,allobjectswithinanSNwereidentical(e.g.,identicalpropertycont).Basedon
this,wesimulatedtheworst‐casescenariowithrespecttoperformance:everyobjectbeing
connectedwitheveryotherobjectineachSNyielding40,4,900and499,000relations.Note
thatonly“issimilarto”and“isauthorof”relationsweredetectedsincethepropertytitleisnot
explicitinplaintextfilesand,therefore,no“hassametitleas”relationswererecognized.
(a)
buzz, cont,
deg, mdate
(c)
cdate, source, uri
(b)
title
(d)
file type
(a) objects
(a)
mdate, weight
(c)
cdate, destination,
vertex, label,
source vertex, uri
(b)
-
(d)
reason
(b) relations
18
Figure12:SNinstancesfortechnicalvalidation
BasedontheinitialSNs,weperformedoperations(add,update,delete)thatrefertoasingle
objectusingthePush‐andPull‐Algorithm(cf.Section3.3).EachcombinationofSN,algorithm
andoperationrepresentedonecasetobeexaminedwithregardtosmallandlargefiles,which
resultedinatotalof36cases.Anexemplarycasemaybeupdatinganobjectwithasizeof1KB
usingthePush‐AlgorithminanSNwith5objects.Foreachcase,thenumbersshowninthe
diagrams(cf.Figs.13‐15)correspondtoaveragesoverthreewarmruns.Warmrunswere
chosentoensurecomparabilityofthemeasuredvaluessincetheiQserGINServer(Wurzer,
2008)performsseveralinitialbackgroundtasksonstart‐up.Notethelogarithmicscaleusedin
thediagrams.
Figure13showsthatobjectscanbeaddedtoanSNinlineartime.Further,depthhasan
influenceontheruntimeregardingthenumberofobjectsinanSNaswellastheirsize(1KBvs.
100KB).Therefore,theactualperformanceofthealgorithmsdependsonthepropertyvaluesof
thevertices.Forexample,thevalueofcont(i.e.,thesizeofthepropertyinbytes)affectsthe
analysisofsimilarityrelationsbetweentheaddedandtheexistingvertices.Detectingrelations
betweentheaddedandtheexistingobjectsispolynomialinthenumberofvertices.Notethat
complexityriseswhenadditionalrelationlabels(m=#ofrelationtypes)mustbedetected
betweenobjects(n=#ofobjects)since,usually,additionalalgorithmsmustbeexecuted,which
examinepropertiesofverticesconcerningaspecificrelationtype.Therefore,thecomplexity
levelcanreachnm^2consideringthateachrelationlabelisderivedbyaparticularalgorithm
whosecomplexityinfluencestheoverallalgorithmiccomplexityaswell.Forexample,ifsuchan
algorithmisexponential,theoverallcomplexitywillnolongerbepolynomial.However,some
algorithmscanbeusedtoprocessmultiplerelationlabels.Asexample,considerderivingforeign
keyrelationshipsfromarelationaldatabase(e.g.,“isauthorof“or“isusedby").
19
Figure13:Effectofdepthandbreadthonadditions
AsshowninFigure14updateoperationsperformsimilarlycompoundtoaddoperations.As
opposedtotheintegrationofobjects,however,certainoperationsarenotrequired(e.g.,for
mandatory/staticproperties).Comparisonsbetweenexistingpropertiesneedtobeperformed
instead,e.g.,tocheckwhetherapropertyhaschanged.However,wecouldnotdetect
significantdifferencesconcerningcostsbetweenaddandupdateoperationswhenapplying
themunderidenticalinitialsituations.
1
10
100
1.000
10.000
100.000
5 5 5 5 50 50 50 50 500 500 500 500
push pull push pull push pull push pull push pull push pull
1 KB 1 KB 100 KB 100 KB 1 KB 1 KB 100 KB 100 KB 1 KB 1 KB 100 KB 100 KB
time in ms
# obj.
alg.
filesize
relation
object
20
Figure14:Effectofdepthandbreadthonupdates
Asopposedtoaddandupdateoperations,deleteoperationsperformdifferently.Whilethe
objectscanbedeletedinlineartime,thedeletionofrelationsvariessignificantlydependingon
thesizeoftheobjects.Thereasonisthatallreferences,whichformthebasisofarelation(e.g.,
extractedconceptsandco‐occurrencesofaspecificobjectincaseof“issimilarto”relations)
mustbedeletedaswell.Notethatsuch“housekeeping”taskswerecontrolledbytheiQserGIN
Server.Inturn,thismightbethereasonforvariationsofthemeasuredvalues.Forexample,the
costfordeletinganobjectwithasizeof1KBoutof5objectswashigherthanthecostfor
deletinganobjectwithasizeof100KBoutof5objects(cf.Fig.15).Furthermore,
measurementswiththeprogramminglanguageJavacanbelessaccuratecomparedtonative
programminglanguages(e.g.,theJavagarbagecollectorcannotbedisabled).
Despitethefactthattheimplementedtechnologymightcauseinaccuraciesinmeasurements,
wewerestillabletoverifythatthemaintenancecostscausedbythealgorithmshighlydepend
ondepthandbreadth(RQ2).However,externalcomponents(e.g.,acomponentforthe
semanticanalysisofobjectproperties)aswellastheircomputationcostcaninfluencethe
performanceofmaintenanceoperationssignificantly.Inturn,theconsiderationofthe
presentedpropertyclassification(cf.Section3.2)canhavepositiveeffectsonperformanceif
onlyrequiredoperations(e.g.,onlyupdatingpropertieswhicharenotmandatory/static)are
executed.
1
10
100
1.000
10.000
100.000
5 5 5 5 50 50 50 50 500 500 500 500
push pull push pull push pull push pull push pull push pull
1 KB 1 KB 100 KB 100 KB 1 KB 1 KB 100 KB 100 KB 1 KB 1 KB 100 KB 100 KB
time in ms
# obj.
alg.
filesize
relation
object
21
Figure15:Effectofdepthandbreadthondeletes
Altogether,thePush‐andthePull‐AlgorithmbothensureasynchronizedSNregardingaffected
datasources.Thereby,thetechnicalvalidationhasshownthatautomaticSNmaintenanceis
feasible(RQ1).Wefurthershowedthattheruntimeofthemaintenancealgorithmsisinfluenced
bydepthandbreadth(RQ2).Inthefollowing,weevaluatetherelevanceaswellasapplicability
ofautomaticSNmaintenanceinthecontextofacasestudyweconductedintheautomotive
domain.
4.3 EmpiricalValidation
Aftercompletingthesuccessfultechnicalvalidation,wehavetoconfirmtherelevanceand
applicabilityofthealgorithmsinareal‐worldusecase(RQ3).FollowingtheproposalofYin
(2009)andKitchenhametal.(1995),wechoseanempiricalcasestudytoevaluateRQ3ina
characteristicprojectsetting.Datawascollectedbysemi‐structuredinterviews,whichallowed
ustoaskopenaswellasclosedquestions(cf.Fig.16).Theinterviewswerebasedona
questionnairecomprising60questions.8Thelayoutofeachinterviewwasdesignedapplyingthe
time‐glassprinciple(RunesonandHöst,2009).Westartedwithclosed‐endedquestionsallowing
forcommentsfromtheparticipantsinthefirstpartwhereweaskedgeneralquestionsabout
theircurrentworkenvironmentandinformationhandlinginthecontextoftheirprocesses.
Afterwardstheyhadtoperformtaskswiththedescribedprototype(cf.Section4.1)andanswer
mostlyclosed‐endedquestionsboundtothesetaskswithrespecttomaintenance(i.e.,add
objects,updateobjects,deleteobjects,andsearchforobjectsandvalidatepropertyvaluesas
wellasdetectedrelations).Finally,weaskedadditional,mostlyopen‐endedquestions.This
approachallowedustodevelopandguidetheinterviewconversationtogaindeeperinsights
throughexplorationaswellasthecollectionofstructureddatabasedonclosedquestions.
1
10
100
1.000
10.000
5 5 5 5 50 50 50 50 500 500 500 500
push pull push pull push pull push pull push pull push pull
1 KB 1 KB 100 KB 100 KB 1 KB 1 KB 100 KB 100 KB 1 KB 1 KB 100 KB 100 KB
time in ms
# obj.
alg.
filesize
relation
object
22
Particularly,inthelastpartoftheinterviews,theopenquestionsallowedreflectingthetasks
withtheSNtogetherwiththeparticipants.Inaddition,theapproachallowedtofurther
understandtheneedaswellastherequirementsforSNmaintenanceandofferedthebasisto
discussthebenefitoffurtherapplicationscenarios.
Figure16:Interviewdesign
Therefore,RQ1couldbeaddressedfromtheuser’sperspectivebythetask‐specificquestions,
whereasRQ3wasinvestigatedbyintroducingandfinalquestions.Theparticipantswere
selectedbasedontheirexpertknowledgeregardingtheconsideredcase.Twobasicroleswere
involved:knowledgeworkersanddecisionmakersfromseveralinnovationdepartmentsinthe
automotivedomain.Thus,allparticipantswereinvolvedinknowledge‐intensivebusiness
processes(MundbrodandReichert,2014).Noparticipantwasamemberoftheresearchteam.
ThecasestudywasperformedinJuly2014withfivedecisionmakersandsixknowledge
workersfromalargeautomotivemanufacturer.Eachinterviewlastedaround90minutes.For
thetasksweprovidedanSNwithidenticalconfigurationregardingthetechnicalevaluation(cf.
Section4.2).However,theunderlyingcorpus(datasources)contained333real‐world
documentsfromtheirfieldofinterest(e.g.,scientificpapersfromdepartmentsdealingwith
technologymonitoringandtechnologydevelopment).Therefore,theuserswerefamiliarwith
theinformationrepresentedintheSNandabletojudgethecontaininginformationandits
qualityonacertainlevel.
23
Figure17:Informationaccessbysource
Theinitialquestionsaboutinformationhandlingintheusers’currentworkenvironment
revealedthat,exceptpersonalcontacts(e.g.,inmeetingsorphonecalls),informationismostly
handledandaccessedinadigitalway(cf.Figs.17and18).
Figure18:Informationaccessbysoftware
However,informationismostlynotwellstructuredandoftendistributedacrossdifferent
0% 20% 40% 60% 80% 100%
Databases/Information Systems
Shared Data Storages
Local Data Storages
Optical Data Storages
Internet
Archives
Non-eletronic media
Colleagues
Other
None
Question: Where do you access information needed for your
daily business?
Knowledge Worker
Decision Maker
Total
0% 20% 40% 60% 80% 100%
File Browser
Intranet
Internet
Individual Software
Manual
Other
Question: How do you access information needed for your daily
business?
Knowledge Worker
Decision Maker
Total
24
sources(e.g.,informationsystemsorshareddrives),whichmakesitdifficulttosearchforit.
Thesestatementsareendorsedbythefactthatasignificantamountofinformationisstoredin
files(cf.Fig.19).Remarkableinthiscontextistheusageofvisualinformation.Almostfifty
percentoftheparticipantsworkwithvisualinformation.Figures,whichmightbeofinterestfor
decisionmakersandknowledgeworkers,areoftenembeddedinfiles(e.g.,chartsin
spreadsheets,modelsinpresentations,ortechnicaldrawingsinpatents)orinformationsystems
(e.g.,reportsordashboards).Theparticipantsstatedthatsuchdrawingsusuallyhadadditional
businessvalue.
Figure19:Informationaccessbyfileformat
Consideringtheenormousfileusage,aresultingchallengewhenworkingwithfilesinabusiness
environmentistokeeptrackofchanges(cf.Fig.20)andtoidentifyinterdependenciessuchas
identicalorsimilardocumentswithindifferentdatasources(cf.Fig.21).Thisismainly
substantiatedbythedynamicsandtheamountofavailableinformationaswellasalackof
technologicalassistance(e.g.,searchoverdifferentdatasources,notificationsonchanged
content).Therefore,inordertobeup‐to‐date,anSNmustcoversuchchanges.
0% 20% 40% 60% 80% 100%
Images
PDF Documents
Text formats
Presentations
Spreadsheets
Audio formats
Video formats
Web formats
Technical formats
Non-electronic media
Other
Question: In which file formats is the relevant information for
your daily business usually stored?
Knowledge Worker
Decision Maker
Total
25
Figure20:Informationdynamics
Inordertoaddressthesechallenges(e.g.,dynamicsofinformationindistributedsources,
heterogeneousfiletypes),weintroducedtheimplementedprototypeandaskedthe
participantstoperformtaskswiththeprovidedgraphicaluserinterface(e.g.,tocheckwhether
propertyvalueswereupdatedortovalidatethecorrectnessofarelation).
Figure21:Informationoverview
Regardingtheconcludingquestions,everyparticipantstatedthatallSNmaintenancetaskswith
theprototypecouldbecompletedsuccessfully.Sinceobjects,relationsandpropertieswere
adaptedbasedonexogenousandendogenouschanges,wecanconfirmthatautomaticSN
maintenanceisfeasibleinpractice.
1
2
3
4
5
lower quartile
minimum
median
maximum
upper quartile
1
2
3
4
5lower quartile
minimum
median
maximum
upper quartile
Statement: The information needed for my daily
business is very dynamic.
1: completely disagree, 2: disagree, 3: neutral,
4: agree, 5: completely agree
Statement: Regarding my current work environment,
I can easily identify interdependencies between
information from different data sources.
1: completely disagree, 2: disagree, 3: neutral,
4: agree, 5: completely agree
26
Figure22:SimplicityofSNmaintenanceforusers
However,certainuserscouldnotrecognizeanyrelationbetweenthedocumenttheyaddedto
theSNandothers.Whileno“isauthorof”and“hassametitleas”relationexisted,the“is
similarto”relationswerenotshown.Toenableamoresophisticateduserexperience,we
filteredtheserelationsaccordingtoahard‐codedweight‐thresholdwithrespecttotheoverall
amountof“issimilarto”relations.Nevertheless,withcorrespondingdatabaseentrieswecould
verifythattherelationsweresetandtookthisasaninputforfurtheruserinterface
developmentbyallowinguserstosetthethresholddynamically(i.e.,filteringrelationsby
weight).
Additionally,weaskedtheusersabouttheirpreferencesforpushorpull.Alluserspreferred
thePush‐Algorithmsincetheeffectsoftheirtaskswerereflectedimmediatelyintheuser
interface.Bycontrast,thePull‐Algorithmalwayshasadelaysinceitisonlytriggeredatacertain
pointintime.Obviously,asynchronizationeverythreeminutescausedtoomuchdelayforsome
users,evenwhenperformingothertasksinbetween.However,thePartial‐Pull‐Algorithm,which
addressesthisdownside,receivedpositivefeedbackfromtheparticipants.NotethatthePartial‐
PullAlgorithmonlyconsidersobjectscurrentlyexistinginanSN.
1
2
3
4
5
Pull/Add Push/Add Pull/Delete Push/Delete
Statement: Updating the SN was easy.
1: completely disagree, 2: disagree, 3: neutral,
4: agree, 5: completely agree
lower quartile
minimum
median
maximum
upper quartile
27
Figure23:SNmaintenanceeffortforusers
WeinterpretthedisadvantageofthePull‐AlgorithmasaconfirmationofRQ3(automatic
maintenanceisessential)andrecommendthePush‐Algorithmiftechnologypermits.Regarding
thePull‐Algorithm,theupdateintervalscanbeshortened,dependingonthenumberoffilesina
datasource.Nevertheless,thedelayofthePull‐Algorithm,asobservedbytheusers,iscaused
bytheconfiguredtimeintervalofimplementedschedulers.However,itsperformance
comparedtothePush‐Algorithmisalmostidentical(cf.Section4.2).
Finally,weaskedtheusersabouttheirimpressionofbenefitsfortheirdailyworkwithrespect
toSNmaintenance.Allusersstatedthatintegratingandconnectingdistributedinformationwas
easywhenusingtheprototype(cf.Fig.22)andcouldbedonewithreasonableeffort(cf.Fig.
23).Additionallytheyconfirmedthebenefitsresultingfortheirdailywork(cf.Fig.24).
Figure24:UserperspectiveonautomaticSNmaintenance
Morethan90%oftheparticipantsstatedthattheSNprovidedanenhancedoverviewon
informationandthatamaintainedSNwasdesirable.Inparticular,thiswouldbeabenefit
comparedtodistributedheterogeneousdatasources.Oneusersummarized:“Maintenanceand
1
2
3
4
5
Pull/Add Push/Add
Statement: My effort updating the SN was low.
1: completely disagree, 2: disagree, 3: neutral,
4: agree, 5: completely agree
lower quartile
minimum
median
maximum
upper quartile
1
2
3
4
5
Pull/Update Push/Update
Statement: Automatic updates of objects, relations
and properties are useful.
1: completely disagree, 2: disagree, 3: neutral,
4: agree, 5: completely agree
lower quartile
minimum
median
maximum
upper quartile
28
correspondingupdatesshouldbeautomatic.Nouserinteractionformaintenanceshouldbe
required.UsersshouldfocusonworkingwiththeSN.”Askingaboutusecasesforamaintained
SN,knowledgeworkersaswellasdecisionmakersrecognizedthepotentialoftheSNtosupport
theirdailywork.
4.4 Conclusion
WehaveshownthatautomaticSNmaintenancewithregardtobothexogenousand
endogenouschangesisfeasiblewithacceptablecost(RQ1).Furthermore,weexaminedthe
effectofdepthandbreadthontheruntimeoftheproposedalgorithms(RQ2).Bothalgorithms
performedsatisfactorilywhenadding,updating,anddeletingobjects.Thecostofdetecting
relations,however,variessignificantlywiththealgorithmsusedforthistask(e.g.,expensive
linguisticorstatisticalalgorithms).
Inthecasestudy,allusersappreciatedaunifiedsinglepointofinformationaccess,whichisup‐
to‐dateandallowsforsearchingandfilteringdistributedinformation.Manyoftheinterviewed
usersalreadyexperiencedawellmaintainedSNasanenablerforvarioususecases(e.g.,expert
searchorgapanalysis).Inparticular,knowledgeworkersanddecisionmakerscanbenefitfrom
maintainedSNs.Bothgroupsemphasizedthatrelationsbetweenobjectscouldbeinteresting
forvariouspurposes(e.g.,navigationwithinanSNordecisionsupport).Forexample,SN
visualizationscanbeusedtosupportdecisions(cf.Section5.2).
Thecasestudyresultsonthedeficitsofinformationhandlingincurrentbusinessenvironments
(cf.Section4.3)haveshownthatitbecomesincreasinglycrucialtoprovideup‐to‐date,
integrated,andhomogenousviewsonenterpriseinformation.Moreover,theempirical
validationconfirmsthatthereisahighdemandforacentralpointofinformationaccessin
knowledge‐intensiveprocesses.Finally,weverifiedthatamaintainedSNisnotonlyessentialfor
suchbusinessprocesses,butapracticablesolutionaswell(RQ3).
5 UseCases
ComplementarytothevalidationofthealgorithmsinSection4,wenowdemonstratethe
relevanceofSNmaintenanceinpracticalsettingsbyconsideringtwousecasesfromthe
technologyprocessintheautomotivedomain.
Radicalinnovationsrequirehighinvestmentsandconscientiouspreparation(Oerteltand
Ulmschneider,2013).Thetechnologyprocessofanenterprise(cf.Fig.25)constitutesthebasis
fordevelopingandadoptingnoveltechnologies,productsandprocessesinastructuredway
(Oertelt,2009)inordertocreatevalueforcustomerswithchangingdemands(e.g.,technical
andfunctionalinnovations,environmentalawareness)inadynamicenvironment(e.g.,volatile
rawmaterialprices,legalregulations,competitors,orstart‐ups).Thetechnologyprocess
involvestasksliketheidentification,monitoring,prioritization,andcontrollingoftechnologies
(OerteltandUlmschneider,2013).Typicaltasks,forexample,includetheobservationof
relevanttechnologies,theidentificationoftechnologyenablers,andtheevaluationofa
technology'smaturitytodeterminefutureapplicationscenariosandevaluatethemwithregard
tocosts,potentialsandrisks.Theoutputofsuchtasksmustbewell‐groundedsince
correspondingdecisionsmighthavesignificantfutureimpactintermsoftechnologicaland
29
economicsuccessaswellasacompany'sreputation.Theprocessrequiressustainableresearch
activities,interactionandexpertiseandtherefore,canbecharacterizedasknowledge‐intensive.
Figure25:Technologyprocess
Inconsequence,identifyingandutilizingrelevantinformationtosupportthetechnology
processconstitutesachallengingtask.Usuallytherequiredinformationcannotbefoundinone
datasourceandisstoredinun‐orsemi‐structureddataformats.Takingtechnologymonitoring
anddecisionsupportasexamples,weillustratehowamaintainedSNcouldreduceeffortsand
enhancethetechnologyprocess.
5.1 UseCase:TechnologyMonitoring
Technologymonitoringcanhelpcompaniestodetectscientific,technical,orsocio‐economic
eventsthathaveanimpactonacompany'sbusiness(BradfordAshtonetal.,1994).Process
tasksincludethemonitoringof
patentpublicationsintheareaofaspecifictopicofinterest,
scientificpublications(e.g.,conferencepapers,dissertations,books),and
competitors(e.g.,analysisoftechnologyusageannouncements)
Usually,suchinformationhasalreadybeenavailableinacompany,butisphysicallystoredat
differentplaces.Consideranengineerperformingasearchonaspecifictopic.Assumethathe
discoversaninterestingpublication,whichcansupportuserswhensolvingatechnicalproblem.
Theengineersavesthedocumentonashareddriveandforwardsthedocumenttoacolleague
fromanotherdepartment.Anotherexampleisatechnologyanalystlookingforinformationfor
aparticularfieldofinterest.Iftheresearchresultisunsatisfactory,atechnologyscoutcanbe
hiredinordertomonitorsignificanttechnologycenters,i.e.,concentrateddomicilesofhigh‐
techcorporations.Tasksofatechnologyscoutincludevisitingfairsorinterrogatingthepersonal
network.Resultsareusuallyreportedbyemailinformofadocumentorapresentation.
Altogether,asillustratedwiththeexamples,informationis,inthecontextofthetechnology
process,typicallygatheredandstoredbyknowledgeworkersatdifferentlocationsinvarious
dataformats.Importantinformationmightnotreacheverystakeholderandtheanalyisofall
availableinformationischallenging.
30
Process Information Information Objects Process Objects
Figure26:SNobjectrepresentationofthetechnologyprocess
TheSNapproachclosesthisgapbyintegratingsiloed(technology)information,bringingitinto
thecontextofprocessandotherinformationobjects(cf.Fig.26),andkeepingitup‐to‐date.
Therefore,inafirststep,theprocessschema(cf.Fig.25)isintegrated(Michelbergeretal.,
2012b).Processschemaelementssuchastasks,events,dataobjects,roles,sequenceflows,
messageflows,associations,orgatewaysarefirstidentifiedandthenusedtocreatethefirst
stageoftheSN(R1andR2,T1‐T6inFig.26).Afterwards,processinstances(e.g.,requestfor
information,decisionfortechnologyusage,allocationoffunds)oftheintegratedprocess
schemasareincludedaswell.Besidestheprocesselements,correspondingmetadata,suchas
author,creationdate,ormodificationdatearealsoconsideredandassociatedwiththeprocess
elements.Certainmetadataisautomaticallyavailable(e.g.,creationdate,modificationdate,
uniqueidentifier),whereasothermetadatahastobedefinedmanually(e.g.,deadline,project
milestone,temporalprocessconstraintorqualitygate).
Inasecondstep,processinformationschemas(e.g.,technologyfactsheettemplatesor
questionnairesfortechnologyevaluation)andprocessinformationinstances(e.g.,technology
factsheetsorcompletedtechnologyevaluations)areintegratedintheSN.Datasourcescanbe
foldersonashareddrivecontainingtechnologyinformation(D1andD2inFig.26;e.g.,pdfor
officedocuments),asuppliermanagementsystem(D3),apatentdatabase(D4),oraproject
managementsystem(D5).Metadata,suchasauthor,filetypeorrevisionnumberisattachedto
informationobjectsaccordingly.
InthethirdstepexplicitrelationsareidentifiedandintegratedintheSN.Examplesincludethe
transformationofrelationsbetweenprocessobjects(e.g.,sequences,roleassociationsderived
fromBPMNmodels)orforeignkeyrelations(e.g.,authorortechnologyfactsheetderivedfrom
arelationaldatabase).Thelastintegrationstepisthedetectionofimplicitrelations(e.g.,
similarityrelationsbetweeninformationobjects).
31
Overall,thementionedstepsoftheillustratedusecaseresultinthepropertyclassification(cf.
Section3.2)andconfigurationoftheSNasshowninFigure27.
Figure27:Propertyclassificationforthetechnologyprocess
ComparedtotheclassificationinSection4.1,weaddedadditionalproperties.Thestatus
property(cf.Fig.27(a))containsthecategorizationofaninformationobjectbasedonitstype.
Thestatuspropertycan,forexample,beusedtostorethelegalstatusofapatent(e.g.,applied,
granted),thematuritystatusofatechnology(e.g.,research,development,ready)orthe
progressstatusofongoingproductdevelopments(e.g.,open,started,finished,approved).In
turn,processobjectsmayusethispropertytostoretheirprocessstatus(e.g.,open,warning,
problem,canceled).Notethatstatuschangesusuallytriggerfurtherprocesssteps(e.g.,
'evaluatetheapplicationofatechnology'whenthestatusischangedtoready).
Optionalandstaticpropertiesareextendedbyrevisionnumberandcontenttype.Revision
numberisusedfortrackingchangesofprocessandinformationobjectswhereascontenttype
specifieswhetheranobjectcontainsinformationaboutatechnology,patent,productor
supplier.Inmostcases,therequiredinformationcanbedirectlygatheredfromthedatasource
(e.g.,patentdatabaseortechnologyfactsheet).
Thetitleisusedtostorenames,e.g.,ataskname,atechnologyname,aproductacronym,a
scientificpublication,apatent,orasuppliername.Usually,thetitledoesnotchangeveryoften
butcanchangeovertime.Sincetitlemightnotbeavailableforeveryobject,itisclassifiedas
optional/dynamic.Furtheroptional/dynamicpropertiesareabstract,description(e.g.,problem
statement,possiblesolutions),chancesandriskswhichallowusersprovidingfurtherdetails
aboutprocessandinformationobjects.Propertiesstartdateandenddate,inturn,store
temporalinformationonaprocessorinformationobject.Forexample,inthecontextofa
processobject,theydefinewhenataskmustbestartedorcompleted.Inthecontextof
informationobjects,thepropertiescanspecifiythereportdate,theestimatedmaturitydateof
atechnologyaswellastheproductionperiodofaproduct.Furthertheycanstorepatent
applicationandexpirationdates.Notethatsuchdatesallowdetectingcomplextemporal
dependencies,e.g.,whetheratechnologyismaturebeforetheproductionofanewproduct
begins(cf.Section5.2).
Analogously,weaddpropertiesstartandenddateforrelations(cf.Fig.27(b))andusethemfor
temporalrestrictionstotriggerendogenouschanges.Examplesincludepatentexpirations,
temporaryroles(e.g.,vacationreplacement)orsupplierchanges.
(a)
buzz, cont, deg,
mdate, status
(c)
cdate, source, uri
(b)
abstract, chances,
description, end
date, risks, start
date, title
(d)
content type, file
type, revision
number
(a) objects
(a)
mdate, weight
(c)
cdate, destination,
vertex, label, source
vertex, uri
(b)
end date,
start date
(d)
reason
(b) relations
32
Furthermore,weenrichtheSNwithrelationlabelslike“transfersinto”or“hasinvented”,
whichwederivefromthebusinessprocessinstances(e.g,.technologytransfersintoaproduct)
andavailableprocessinformation(e.g.,patentinventor).Asexample,transferinformationor
inventorscanbederivedfromtechnologyfactsheetsandtransformedtorelationsaccordingly.
Inventorsofapatentcanbedirectlyderivedandprocessedfromapatentdatabasewhereas
productsusuallydonothaveaninventor.Notethattheadditionalsemanticsofrelationswillbe
thebasisforfurtherapplicationscenarios(cf.Section5.2).
Afterconfiguration,theSNisreadyforuse.Thefirststepofthetechnologyprocess,the
technologyidentification,canbesupportedbyintegrating,structuringandinterrelating
informationfromunderlyingdatasources(cf.Section2).Examplesoftechnologyrelated
artifactsareillustratedinTable2.Knowledgeworkersanddecisionmakersthenbenefitfroma
unifiedsinglepointofinformationaccesswheretheycansearchandfilteravailabletechnology
information.Consideranengineerlookingforinformationaboutautonomousdriving.Hemight
beabletofindanewapproachwhichiscurrentlyevaluatedbyanotherdepartment.Further,he
mayfollowrelatedinformationobjectsanddetectsascientificpublicationdescribinganovel
collisionavoidancetechnology.Anotherengineercan,startingwithatechnologyofinterest,
identifytechnologyalternativesbyfilteringrelatedtechnologieswhichareconnectedby”is
similarto”relations.Notethatrelatedinformationobjectscanprovidehintsforthesolutionof
atechnicalproblemaswell.
Thesecondstep,technologymonitoring,canbesupportedbySNmaintenance,i.e.,byadapting
endogenousandexogenouschanges(cf.Section3.1).Considerknowledgeworkersanddecision
makersfromdifferentteamswhoarecontinuouslyupdatingtechnologyinformationin
documentsorenterpriseinformationsystems.Amajorbreakthrough,suchasasolutionfora
specifictechnicalproblem,thediscoveryorthematurityofatechnology,isusuallyfiled,but
mightnotreachallknowledgeworkerswhohadinterest.TheSNcapturessuchchanges
immediatelyandsendsnotificationsaccordingly(i.e.,byroleorauthorassociationsofsimilar
objects).Decisionmakersreceivenotificationswhenanewmaturitylevelofatechnologyis
reachedoradditionaleconomicinformation,e.g.,oncost,becomesavailable.
Therefore,theSNcapturesexogenousandendogenouschanges,i.e.,ifnewinformation(e.g.,a
patentorinformationaboutatechnologysupplier)becomesavailableorobjectsandrelations
needfurtherprocessing(e.g.,arelationorinformationobjectisoutdated).Basedonsuch
changes,theSNisabletoinitiatefurtherprocessstepsbysendingnotifications.
5.2 UseCase:DecisionSupport
Buildingontheprevioususecase,wedemonstratehowtechnologyanalysiscanbeenhanced
throughSNmaintenance.
Increasedmarketdynamics,changingcustomerdemands,extendedproduct‐portfolios,
multiplestakeholdersoruncertainty(e.g.,aboutconsequencesofadecision)areexamplesof
challengeswhichexecutiveshavetocopewithinenterpriseswhenmakingdecisions.Such
factorstypicallyincreasethecomplexityofdecisions,makingitdifficulttodetermineanddecide
forthebestoption.Asexample,abroaderproductrangeandvarietyonproductderivates
requireshigherdensityofdecision‐makingwithprevisionanddeeperunderstandingof
expectedinterdependencies(e.g.,complexityofinterdependenciesofproductsandtheir
33
components).Usually,manytasksrelatedtothedecisionprocessareperformedmanually
whichcanbeverytime‐consuming.Asexample,considerthepreparationofastrategic
roadmapforamanagementmeetingortherequestforinformationfromprocessstakeholders.
Thus,providingup‐to‐dateviewsonintegratedandcross‐linkedinformationbyrequestcanbea
businessbenefit,e.g.,toreducepreparationtimefordecisionmakers.
Asexample,adecisionmakerhastoconsiderthematuritylevelofatechnologyandcompare
cost,benefit,qualityandavailabilitywithotheralternatives–ifavailable.Examplesofgeneral
questionsareasfollows:Doesthetechnologyfitintotheproductportfolio?Whenwillthe
technologybeavailable?Canthetechnologybeimplementedwithreasonablecost?
Therefore,theon‐demandavailabilityofanintegratedviewonallinformationrequiredfor
analysisisdesirable(cf.Section4.3).
TheSNsupportsthissub‐processbyunveilingalternativesaswellaspossibleconnectionsto
otherrelatedinformationobjects(cf.Tab.2)throughsemanticrelations.Inordertodetermine
atechnology'smaturity(e.g.,status)aswellasforcapturingtemporalfacts(e.g.,startdate,end
date),weusetheconfiguredpropertiespresentedinSection5.1.
Artifact Characteristics
Fact sheets Strategy, topic field, idea, technology, project, material, patent,
competence, product, production technique
Guidelines and
templates
Checklist, order, request form (e.g., for funds)
Scientific
publications
Scientific paper, master thesis, dissertation, survey, journal article
Reports and
protocols
Meeting and workshop protocol, decision protocol, fair report, trend
report, customer feedback, internal research and laboratory reports,
competitive and market analysis, press release, reverse engineering report
Lessons learned
and best practices
Project documentation, workshop preparation, strategies and creativity
techniques
Table2:Technologyrelatedartifacts
TheadditionalpropertiesallowustovisualizetheSNanditsunderlyingobjectsinaflexible
way,i.e.,differentviewsontheSNcanbegenerated.OneSNviewsupportingthedecision
processisastrategicroadmap(cf.Fig.28).Adecisionmakercanspottemporalaspectsof
technologies,likeexpectedmaturitydates,whenarrangingtechnologyobjectsonatimeline.
Additionally,relatedprocessinformationlikeprojects,patentsormaterialscanbeaddedtothe
view,whichallowsdetectingtimeconstraints(e.g.,whetheratechnologywillbereadyor
whetheratechnologyisstillprotectedbyapatentwhenaprojectrequiresthistechnology).A
searchandfilterfunctionalitylimitstheresulttorelevantobjectsandthereforereducesthe
viewtorelevantinformationbyusingavailableproperties(e.g.,creationdate,contenttype,file
typeorstatus).Evaluations(e.g.,chancesandriskanalysis)fromexpertslikeengineerscan
supportopinionmaking.
Forexample,adecisionmakerfilterstechnologieswithstatus“ready”andproduct
developmentprojectswhichhavenotstartedyetandonlyselects“transfersinto”relations.By
arrangingtheseobjectsandrelationsonthex‐axiswithtemporalinformation,usingstartand
enddates,criticalproductdevelopmentprocesses(incaseatechnologymightnotbemature)
34
canbeidentified.Furthermore,thevisualizationcanbetriggeredandadaptedondemand
avoidingmanualpreparationwork.
Figure28:Strategicroadmap(datamasked)
SuchviewsontheSNarenotlimitedtotime‐basedcharts.Consideraportfoliochartwhich
groupstechnologiesbytheirmaturity,expectedcostsavingsorfieldsofinterest(e.g.,
autonomousdriving).
Therefore,SNmaintenancedoesnotonlyallowfortheintegrationandglobalsearchof
technologyinformation,butenablesreal‐timeanalysisaswell.
Summingup,amaintainedSNsupportsthetechnologyprocessbyprovidingaccesstoup‐to‐
datetechnologyinformationinanintegrated,connectedandsearchableway.Processsteps,like
technologyanalysis,technologymonitoring,technologycontrollingandsearchforalternatives,
canbesupported.Relatedtechnologiescanbeidentifiedandprocessstepsbetriggeredby
capturingendogenousandexogenouschanges,e.g.,throughnotificationsincaseofanew
technologyoranupdatedmaturitylevel.Furthermore,decisionmakerscanbesupportedby
deliveringinformationthroughadequateviewsondemand.
Infuturework,wewillsupportfurtherprocessstepsinordertoallowrationalizingdecisionsby
socializinginformationintermsofevaluation(e.g.,capturingopinionsofengineersona
technology'smaturityinamoreprofoundway)andenrichinternalinformationobjectswith
informationfromtheWorldWideWeb(e.g.,selectedtechnologyblogs).
6 RelatedWork
Asemanticnetworkrepresentsdomain‐specificknowledgeinastructuredandmachine‐
interpretableway(Reichenberger,2010).Generally,varioustypesofsemanticnetworksexist.
Figure29showsthemostcommonapproaches,i.e.,associativenetworks,topicnetworks,fact
networks,andontologies.Thex‐axisrepresentstheeffortrequiredtocreateasemantic
network,whereasthey‐axisrepresentsthedegreeofsupportasemanticnetworkprovidesfor
35
particularusecasessuchassearchrefinement,semanticsearch,visualization,orreasoning.
AccordingtoReichenberger(2010),wedistinguishbetweenlight‐andheavy‐weightedsemantic
networks.
Figure29:Typesofsemanticnetworks
AsshowninFigure29,theeffortforcreatingassociativenetworks(Findler,1979)aswellas
theirdegreeofsupportarelow.Associativenetworksaremainlyusedforsearchrefinement.In
turn,topicnetworks(ParkandHunting,2002)provideahigherdegreeofsupportthan
associativenetworks,buttheeffortforcreatingthemissignificantlyhigheraswell.Theyare
oftenusedforrealizingasimplenavigationinsemanticnetworksandforvisualizingrelated
topics.Inturn,factnetworks(Reichenberger,2010)provideahigherdegreeofsupport(e.g.,
conceptsaresupported).Theyareusedforrealizingpersonalizedviews(e.g.,context‐aware
search)andnavigationtrees(e.g.,moderatedsearch).Finally,thehighestdegreeofsupportis
offeredbyontologies(Lacy,2005;Stuckenschmidt,2009),whichprovidegoodresultsinrespect
toconceptualizationanddelimitationofconcepts.However,hugemanualeffortisneededto
createhigh‐qualityontologies.Importantusescasesaresemanticsearchandreasoning.
Unlikeexistingsemanticnetworks,anSNfocusesoninformationandprocessobjectsaswellas
theirrelations.LessmanualeffortisneededtocreateormaintainanSN(Wurzer,2008).Note
thatresearchinthefieldofsemanticnetworkshasmainlyfocusedontherepresentationof
domain‐specificknowledgeinastructuredandmachine‐interpretableform.Whathasbeen
neglected,however,isthemaintenanceofsemanticnetworks.Generally,semanticnetworks
haveincommonthattheymustbemaintained.Dependingonthetypeofthesemanticnetwork
(cf.Fig.29),however,thelevelofmaintenanceeffortvarieswidely.Commonly,thehigherthe
efforttocreateasemanticnetworkis,thehigherthemaintenanceeffortwillbe.Semi‐
automaticmaintenanceapproachesareprovided,forexample,byČapek(2009),Gargourietal.
(2003),andDinhetal.(2014).However,theseapproachescannotbedirectlyappliedtotheSN
sincetheydonotallowforentirelyautomatedSNmaintenanceasitisnecessaryinmany
scenarios.
Moreover,inrecentyears,variousapproacheswereproposedtotacklethedeliveryof
informationtousersincludingdatawarehousing(KimballandRoss,2013),businessintelligence
(Kolb,2012),decisionsupportsystems(Sauter,2011),andenterprisecontentmanagement
(RockleyandCooper,2012).Datawarehousing,forexample,ratherfocusesonthecreationof
anintegrateddatabase(Lechtenbörger,2001).Opposedtothis,anSNdealswiththedeliveryof
36
processinformationtosupporttheeffectiveandefficientexecutionofbusinessprocesses.
Traditionalbusinessintelligence,inturn,enablesdataanalysisandisusuallycompletelyisolated
frombusinessprocessexecution(BucherandDinter,2008).Moreover,informationsupplyis
oftenrestrictedtodecisionmakersonexecutivelevel(BaarsandKemper,2008;Rouhanietal.,
2012).Conversely,ourapproachfocusesontheintegrationandanalysisofprocessinformation
aswellasitsdeliverytobothknowledgeworkersanddecisionmakers.Bycontrast,decision
supportsystemssupportdecisionmaking;i.e.,theyservetheexecutivelevel(Janakiramanand
Sarukesi,2004)andareusuallybasedonstructureddata(e.g.,salesfigures).Enterprisecontent
management,inturn,dealswiththemanagementofinformationacrossenterprisesreferringto
strategies,methods,andtools(Cameron,2011).
7 SummaryandOutlook
ThispaperpresentedanapproachforSNmaintenance,i.e.,foradoptingexogenousand
endogenouschanges.Forthispurpose,wepresentedthreealgorithmsbasedontheproperty
classification,whichallowsforanefficientSNmaintenance.
ThevalidationconfirmsthattheSNmaintenanceapproachisabletokeepSNssynchronized
withunderlyingdatasourcesinreasonabletime.Moreover,weappliedthealgorithmstoareal‐
worldcase,i.e.,wevalidatedthembasedonanimplementationintheautomotivedomain.
Furthermore,weillustratedthepracticabilityandusabilitywithtworeal‐worldusecasesfrom
thetechnologyprocess.
Infuture,wewillimprovethealgorithmstosupportself‐learningcomponentswithafocuson
endogenouschanges.Therefore,wewillextendtheinformationextractioncomponentto
retrieveadditionalfacts(e.g.,competitoractivities)whichcanbefurtherprocessedbya
reasoningcomponent.Additionallywewilladdtaxonomicsupportsothatautomatic
maintainanceisstillensured,i.e.,thecomponentwillnotaffectthealgorithms.Theextensions
willallowustooptimizesearchandfiltercapabilities(e.g.,extendedfacetsforfiltering)aswell
asinferringnewfactsfromproperties(e.g.,"competitorlaunchedanewproductinChina"),
whichmayresultinfurtherrelations(e.g.,"isactive"relationtoaninformationobject
describingtheAsianmarkets).Therefore,wewillincreasethedegreeofsupport(cf.Section6)
whilekeepingtheeffortlow.
References
Baars,H.andKemper,H.‐G.(2008),"ManagementSupportwithStructuredandUnstructured
Data‐AnIntegratedBusinessIntelligenceFramework",inJofInformationSystems
Management,Vol.25,No.2,pp.132‐148.
Bernstein,P.A.andHaas,L.M.(2008),"InformationIntegrationintheEnterprise",in
CommunicationsoftheACM,Vol.51,No.9,pp.72‐79.
BradfordAshton,W.,Johnson,A.H.andStacey,G.S.(1994),"Monitoringscienceandtechnology
forcompetitiveadvantage",inCompetitiveIntelligenceReview,Vol.5,No.1,pp.5‐16.
37
BradfordAshton,W.,Kinzey,B.R.andGunn,M.E.(1991),"Astructuredapproachformonitoring
scienceandtechnologydevelopments",inInt'lJofTechnologyManagement,Vol.6,No.1‐2,pp.
91‐111.
Bucher,T.andDinter,B.(2008),"ProcessOrientationofInformationLogistics‐AnEmpirical
AnalysistoAssessBenefits,DesignFactors,andRealizationApproaches",Proc41stHawaiiInt'l
ConfonSystemSciences(HICSS'08),Hawaii,UnitedStates,pp.392‐402.
Cameron,S.A.(2011),EnterpriseContentManagement:ABusinessandTechnicalGuide,1st
edition,BritishInformaticsSociety,Swindon,UnitedKingdom.
Čapek,T.(2009),"SemanticNetworkIntegrityMaintenanceviaHeuristicSemi‐AutomaticTests",
Proc.RASLANWorkshop2009,Brno,CzechRepublic,pp.63‐67.
Dinh,D.,DosReis,J.C.,Pruski,C.,DaSilveira,M.andReynaud‐Delaître,C.(2014),"Identifying
relevantconceptattributestosupportmappingmaintenanceunderontologyevolution",inWeb
Semantics:Science,ServicesandAgentsontheWorldWideWeb,Vol.29,pp.53‐66.
Farwick,M.,Agreiter,B.,Breu,R.,Ryll,S.,Voges,K.andHanschke,I.(2011),"Requirementsfor
AutomatedEnterpriseArchitectureModelMaintenance‐ARequirementsAnalysisbasedona
LiteratureReviewandanExploratorySurvey",Proc.13thInt'lConferenceonEnterprise
InformationSystems(ICEIS'11),Beijing,China,pp.325‐337.
Findler,N.V.(1979),Associativenetworks,1stedition,AcademicPress,NewYork,UnitedStates.
Fischer,R.,Aier,S.andWinter,R.(2007),"AFederatedApproachtoEnterpriseArchitecture
ModelMaintenance",2ndInt'lWorkshoponEnterpriseModellingandInformationSystems
Architectures,St.Goar,Germany,pp.9‐22.
Freund,J.andRücker,B.(2012),Real‐LifeBPMN:UsingBPMN2.0toAnalyze,Improve,and
AutomateProcessesinYourCompany,1stedition,CreateSpaceIndependentPublishing
Platform,NorthCharleston,UnitedStates.
Gargouri,Y.,Lefebvre,B.andMeunier,J.‐G.(2003),"OntologyMaintenanceusingTextual
Analysis",Proc.7thWorldMulticonferenceonSystemics,CyberneticsandInformatics(SCI),
Orlando,UnitedStates,pp.248‐253.
Hipp,M.,Michelberger,B.,Mutschler,B.andReichert,M.(2013),"AFrameworkforthe
IntelligentDeliveryandUser‐adequateVisualizationofProcessInformation",Proc28th
SymposiumonAppliedComputing(SAC'13),Coimbra,Portugal,pp.1383‐1390.
Hipp,M.,Mutschler,B.,Michelberger,B.andReichert,M.(2014),"NavigatinginProcessModel
RepositoriesandEnterpriseProcessInformation",Proc8thInt'lConfonResearchChallengesin
InformationScience(RCIS'14),Marrakesh,Morocco,pp.1‐12.
Hipp,M.,Mutschler,B.andReichert,M.(2011),"OntheContext‐aware,PersonalizedDelivery
ofProcessInformation:Viewpoints,Problems,andRequirements",Proc6thInt'lConfon
Availability,ReliabilityandSecurity(ARES'11),Vienna,Austria,pp.390‐397.
Hotho,A.,Nürnberger,A.andPaaß,G.(2005),"ABriefSurveyofTextMining",inLDVForum‐
GLDVJournalforComputationalLinguisticsandLanguageTechnology,Vol.20,No.1,pp.19‐62.
38
Janakiraman,V.S.andSarukesi,K.(2004),DecisionSupportSystems,1stedition,PrenticeHall,
NewDelhi,India.
Kimball,R.andRoss,M.(2013),TheDataWarehouseToolkit:TheDefinitiveGuideto
DimensionalModeling,3rdedition,JohnWiley&Sons,Indianapolis,UnitedStates.
Kitchenham,B.,Pickard,L.andPfleeger,S.L.(1995),"CaseStudiesforMethodandTool
Evaluation",inIEEESoftware,Vol.12,No.4,pp.52‐62.
Kolb,J.M.(2012),BusinessIntelligenceinPlainLanguage:ApracticalguidetoDataMiningand
BusinessAnalytics,2ndedition,AppliedDataLabs,Plainfield,UnitedStates.
Lacy,L.W.(2005),OWL:RepresentingInformationUsingtheWebOntologyLanguage,1st
edition,TraffordPublishing,Victoria,Canada.
Lechtenbörger,J.(2001),DataWarehouseSchemaDesign,1stedition,Infix,St.Augustin,
Germany.
Lewandowski,D.(2005),"WebSearching,SearchEnginesandInformationRetrieval",in
InformationServices&Use,Vol.18,No.3.
Michelberger,B.,Mutschler,B.,Hipp,M.andReichert,M.(2013),"DeterminingtheLinkand
RatePopularityofEnterpriseProcessInformation",Proc21stInt'lConfonCooperative
InformationSystems(CoopIS'13),Graz,Austria,pp.112‐129.
Michelberger,B.,Mutschler,B.andReichert,M.(2011a),"TowardsProcess‐oriented
InformationLogistics:WhyQualityDimensionsofProcessInformationMatter",Proc4thInt'l
WorkshoponEnterpriseModellingandInformationSystemsArchitectures(EMISA'11),Hamburg,
Germany,pp.107‐120.
Michelberger,B.,Mutschler,B.andReichert,M.(2011b),"OnHandlingProcessInformation:
ResultsfromCaseStudiesandaSurvey",Proc2ndInt'lWorkshoponEmpiricalResearchin
BusinessProcessManagement(ER‐BPM'11),Clermont‐Ferrand,France,pp.333‐344.
Michelberger,B.,Mutschler,B.andReichert,M.(2012a),"AContextFrameworkforProcess‐
OrientedInformationLogistics",Proc15thInt'lConfonBusinessInformationSystems(BIS'12),
Vilnius,Lithuania,pp.260‐271.
Michelberger,B.,Mutschler,B.andReichert,M.(2012b),"Process‐orientedInformation
Logistics:AligningEnterpriseInformationwithBusinessProcesses",Proc16thInt'lEnterprise
ComputingConf(EDOC'12),Beijing,China,pp.21‐30.
Michelberger,B.,Ulmschneider,K.,Glimm,B.,Mutschler,B.andReichert,M.(2014),
"MaintainingSemanticNetworks:ChallengesandAlgorithms",Proc16thInt'lConfon
InformationIntegrationandWeb‐basedApplications&Services(iiWAS'14),Hanoi,Vietnam,pp.
365‐374.
Mundbrod,N.andReichert,M.(2014),"Process‐AwareTaskManagementSupportfor
Knowledge‐IntensiveBusinessProcesses:Findings,Challenges,Requirements",Proc3rdInt'l
WorkshoponAdaptiveCaseManagementandotherNon‐workflowApproachestoBPM
(AdaptiveCM'14),Ulm,Germany,pp.116‐125.
39
Oertelt,S.(2009),Innovationscontrolling‐GanzheitlichesVerfahrenzurPriorisierungund
SteuerungvonVorentwicklungsprojekten,1stedition,Shaker,Magdeburg,Germany.
Oertelt,S.andUlmschneider,K.(2013),"ProzessintegrierterEinsatzvirtuellerMethodenim
strategischenTechnologieundInnovationsmanagement",KnowTech‐Wissensmanagementund
SocialMedia‐MarkterfolgimInnovationswettbewerb,Hanau,Germany,pp.485‐498.
Park,J.andHunting,S.(2002),XMLTopicMaps:CreatingandUsingTopicMapsfortheWeb,1st
edition,Addison‐Wesley,Boston,UnitedStates.
Reichenberger,K.(2010),KompendiumsemantischeNetze:Konzepte,Technologie,
Modellierung,1stedition,Springer,Berlin,Germany.
Reichert,M.andWeber,B.(2012),Enablingflexibilityinprocess‐awareinformationsystems,1st
edition,Springer,Berlin,Germany.
Rockley,A.andCooper,C.(2012),ManagingEnterpriseContent:AUnifiedContentStrategy,2nd
edition,NewRiders,Berkely,UnitedStates.
Rouhani,S.,Asgari,S.andMirhosseini,S.V.(2012),"ReviewStudy:BusinessIntelligence
ConceptsandApproaches",inAmericanJofScientificResearch,Vol.50,pp.62‐75.
Rowley,J.(2007),"Thewisdomhierarchy:representationsoftheDIKWhierarchy",inJof
InformationScience,Vol.33,No.2,pp.163‐180.
Runeson,P.andHöst,M.(2009),"Guidelinesforconductingandreportingcasestudyresearch
insoftwareengineering",inEmpiricalSoftwareEngineering,Vol.14,No.2,pp.131‐164.
Sauter,V.L.(2011),DecisionSupportSystemsforBusinessIntelligence,2ndedition,JohnWiley&
Sons,Hoboken,UnitedStates.
Scheer,A.‐W.(2002),ARIS‐VomGeschäftsprozesszumAnwendungssystem,4thedition,
Springer,Berlin,Germany.
Stuckenschmidt,H.(2009),Ontologien:Konzepte,TechnologienundAnwendungen,1stedition,
Springer,Berlin,Germany.
Wurzer,J.(2008),"NewApproachforSemanticWebbyAutomaticSemantics",Proc2nd
EuropeanSemanticTechnologyConf(ESCT'08),Vienna,Austria.
Yin,R.K.(2009),CaseStudyResearch:DesignandMethods,4thedition,SagePublications,
ThousandOaks,UnitedStates.
[1]http://www.playframework.com/
[2]http://getbootstrap.com/
[3]http://d3js.org/
[4]http://jquery.com/
[5]http://sourceforge.net/directory/?q=nipro
[6]http://lucene.apache.org/
[7]http://www.mysql.com/
[8]Availableathttp://nipro.hs‐weingarten.de/casestudy
