Knowledge Graphs and Machine Learning - Stardog

Knowledge graphs make it easier to feed better and richer data into ML algorithms. They do this by helping you leverage industry-standard models ... KnowledgeGraphsandMachineLearning ColleenLuther Mar17,2022,10minuteread Getthelatestinyourinbox backtoallposts Accordingtothe2020AIinOrganizationsSurvey*,23%oforganizationsdeployedgraphtechniquesintheirartificialintelligence (AI) projects.Largetechnologycompanies,public-sectororganizations,financialsolutionsproviders,andhealthcareledtheway,usinggraphtechnologiestoenhancedatasearch,informationretrieval,andrecommendations.Butnow,allindustriesandorganizationscancombineknowledgegraphswithmachinelearningbyusingplatformsthatareeasytoadoptandscale,makingmachinelearningmorecommonplaceandmoresuccessful.  WhatisaKnowledgeGraph?  Knowledgegraphs connectandcontextualizedisparatedata,organizedandrepresentedingraphdatabases.Builttocapturetheever-changingnatureofknowledge,knowledgegraphseasilyacceptnewdata,datasets,definitions,andrequirements. Howdotheydosuchwonders?Knowledgegraphswerebornfromthe semanticweb,whichwasTimBerners-Lee’sattempttodiscernmeaningfulrelationships (using RDF metadatamodels,goingbeyondthesimplelink)betweeninformationonwebpages.Thesemanticwebvisionhasnotmaterializedbuttheunderlyingtechnologieshaveprovenverysuccessfultoconnectsilosofenterprisedatainwaysthatgivecontexttothedata. *Semantics* isaboutencodingmeaningalongwithdata. So,aknowledgegraphisasemanticdatalayer.Andinthatlayer,knowledgeworkers (often assistedbyinferencing)describehowallthedatathattheorganizationaccessesisclassifiedandrelated.Aknowledgegraphdescribesthemeaningofallthesebusinessobjectsbynetworkingthemandbyaddingtaxonomiesandontologicalknowledgethatprovidescontext.Thisdatalayerprovidesasecureaccesspointthatisstandards-basedandmachine-processable. Graphdatabasesarebuiltforstorage.Graphstructurealone,withouttheinferencing,virtualization,andagiledevelopmentavailableinenterpriseknowledgegraphplatforms,wouldrequireimmenseworktoscaletotheenterpriselevel.Withaknowledgegraph,datascientistscanworkwithknowledgeengineers,togetherwithbusinessusersandinformationtechnologyteams,toturndataintoactionableinsights.Aslarge-scaleaccesstodatacontinuestogrow, theenterprisewillneedtoworktogether. Graphmachinelearningisapowerfultooltohelp. WhatisMachineLearning? Machinelearning (ML) iswhenmachineslearnfromdataandself-improve.In1952,ArthurSamuelcreatedaprogramtohelpanIBMcomputergetbetteratcheckersthemoreitplays,soMLalgorithmshavebeenaroundforover70years.MLiscommonplaceforrecommendations,predictions,andlookingupinformation.  MLisaformofartificialintelligence,whichisawideareaofcomputersciencefocusedonbuildingsmartmachinesthatrequirehumanintelligence.AIstrategygenerallyfocusesonensuringdataisaccessible,reusable,interpretable,andhighquality,whichisoftenachallengewithexistingdatainfrastructure.  TheEvolutionofMachineLearningApproaches Butthingshavechangedsincethe1950s.ApproachingMLandAIisnotstraightforward.TherearetwobranchesofAIoccurringwithinenterprisestoday.Mostorganizationshistoricallyincorporatedstatisticallearningthroughdatascienceprojects.Butrule-basedAIisgrowing,andthisapproachincludeseverythingfrommakingintelligentinferencesaboutschemastoexpeditedataintegrationtoassemblingtechniquesfortextanalyticsorNaturalLanguageProcessing (NLP).OurCEOnotedto AnalyticsWeek: InsteadofthesedifferentbranchesofAIcompetingwitheachotherinvendorsolutions,theindustryhasreachedapointofinflectioninwhichtherearemoreofferings “doing newschoolAI,i.e.,statisticallearning,machinelearning,whatwecallmachinelearningandalso,atthesametime,andwe’veworkedonthis,soitallworkstogetherseamlessly,they’realsodoingthatsymbolicorrules-basedAI,”StardogCEOKendallClarkcommented…. “It looksmoreandmorelikethefutureofAIwillbesomecombinationofbothlogicalandstatistical.” IsaKnowledgeGraphPartofMachineLearning? AknowledgegraphisnotinherentlyapartofML,butitcanhelpyoualot.Thebestdatascienceprojectscomefromcombiningmorethanonesourceofdata,andthatcanbeanightmarefordatascientists.Whenitcomestocombiningdatasourcesanddatasets,ontologiesandcontexthelp.ThismeansplatformslikeStardogassisttremendously.  Therearemanydifferenttoolstochoosefrom,concerningknowledgerepresentation.Andwithsomanyusecasesanddependencies,datapointsanddatasources,successdependsonwhatyoulooktoaccomplish.Forexample,areyourunningdeeplearningtoclassifydatatoturnthatintoaknowledgegraph?Buildinga recommender system?Orexploringneuralnetworks?Areyoucreatingachatbotorcreatinga “Wikipedia” orsearchengineforyourknowledgebase?  Fromanimplementationperspective,themanypossiblepathscanfeellikeabarriertoentryfororganizationsthatjustwanttogetstarted.It’sagoodthingthatknowledgegraphshelparangeofAI/MLapproaches.  WhatisaKnowledgeGraphinMachineLearning? KnowledgegraphsmakeiteasiertofeedbetterandricherdataintoMLalgorithms.Theydothisbyhelpingyouleverageindustry-standardmodelsandontologies,modelyourdomainknowledge,andconnectdisparatedatasourcesacrosstheenterprise.YoucanmaximizetheuseandreuseofyourinternalcontentbylayingthefoundationforAIandsemanticapplications.Ultimately,connectandshowmeaningfulrelationshipsbetweenyourdata,regardlessofstoragearea,size,type,andformat.  Secondarybenefitsinclude: Creation:Youcanusedeeplearningtohelpdeterminethedataclassificationsneededforaknowledgegraph Insights:Classifynodesand/orgroupnodes,andpredictmissingconnections Output:Enhancetheoutputofdata-drivenMLmodelswithknowledgefromthegraph. HowKnowledgeGraphsandMachineLearningWorkTogether  TheinherenttraitsofknowledgegraphspositthemasatoptoolofmodernAIandMLstrategy.Let’sexamineafewwaysinwhichtheyhelp.  Enablehighlyproductivedataworkers Howdodatascientistsandmachinelearningengineersspendtheirtime?Asignificantportionoftheirtimeisspentdatawrangling (also knownasdatamunging).Datawranglingistheprocessofmanualdatagatheringandcleansingbeforeusingit.Itincludesthingsliketryingtofindtherightdatasourceorgettingtheirextractstobuildtheirmatricestofeedtheiralgorithms.Inreal-worldsituations,thistypeofworkcanofteneatup70-80%ofthetimeittakestoproducethedesiredmodelorexpectedresults.  Withaknowledgegraph,datascientistscantrainmodelsdirectlyonunifieddata—withharmonizedterminology andsynthesized datasources—insteadofonincomplete,out-of-date,orinaccuratedata.Soevenatthemostbasiclevel,pullingthedatafromaplatformlikeStardogisahugetimesaver. Nowlet’sconsidertherules-basedapproach.Stardoghasbest-in-classsupportforsomethingcalledinferencing.Stardog’sInferenceEngineallowsyoutoresolveconflictingdatadefinitionswithoutchangingorcopyingtheunderlyingdata.Captureyourbusinessanddomainrulesinthedatamodel;theInferenceEngineintelligentlyappliestheserulesatquerytime.Thiseasilysolvesacommonissue—whatonedatabasecallsa “Major Account”anothercallsan“EnterpriseCustomer.” Tofix,writearulethatstatesbotharesubclassesof “Top Accounts”andquerythefullrangeofdetailsonyourconnecteddata.TheInferenceEnginedisplaysalllogicforeachresult,makingexplainableAIareality. Soagain,ifyourobjectiveistopulloutawell-knowndatasettotraindataon,orit’sgoingtobeyourvalidationdataset,thenaplatformlikeStardogwillmakethoseefficientandhighlyproductiveactivities.  Workseamlesslywithyourexistingtools Knowledgegraphsthatcontainvirtualization (not justgraphdatabasesactingasaknowledgegraph)workwelltomaintaindataaccuracyandthesecurityofexistingtools.Thisincludesincumbenttoolchainsandframeworksthatarealreadydeployed.A knowledgegraphthatisatruedatalayer doesnotrequireyoutochangeanythingyou’redoingtoday. OurplatformreinforcestheproductivityboostbecausethetoolsmatchthoseuserbasesverywellandcomprisethingslikePythonsupport,Rlibraries,etc.Additionally,theoutputofyourmodelscanalsobeputbackintotheknowledgegraph.  Becauseaknowledgegraphisalsoasemanticlayer,itenablesreuseandinteroperability.Thiscreatesanenterprise-wideassetthatdoesnotrequire reinvention foreachindividualapplicationorusecase.Youcansolveinfrastructurechallengeswithouthavingto redo orloseyourexistingsystemsandwithouthavingtobuildeverythingfromscratcheachtime.  Getstarted inAIandmachinelearningquickly  Youneedahigh-performancetoolthatgivesyoufastaccesstothemaximumamountofdataregardlessofwhereit’sstored. Stardog’splatformprovidesagreatwaytoaccessdata—via virtualization,whichleavesdatawhereitis. Thishelpsinsulateyoufromchangesin yoursourcedata.Ifdatachanges,youcanstillquicklyretrainandredeploythemodels.Whenyouconsiderhowtoimprovemodelquality,trainingagainstrawdataisnowacost-effectiveandscalablesolution.  Ourplatformalsoshipswithbuilt-inpredictiveanalyticsandsimilaritysearch,supportingquickmodeldevelopmentanditerationfordataanalysis. It’shelpfultohavetheabilitytopredictnodesandedgesinaknowledgegraph.YoucanuseStardogtoextractpatternsfromyourdataandmakeintelligentpredictionsbasedonthosepatterns.UseMLtopredictthevalueofarelationshiporcombinewithPathfindertosolvetrickyoperationalproblemslikedeterminingthebestalternativesupplyrouteswhenadepotgoesdown.  Similaritysearchisalsoanimportantfunctionalitywhencoupledwiththeconnecteddataofaknowledgegraph,asyoucanuseittodetectandrecommendpatterns.Forexample,usesimilaritysearchtorecommendrelevantarticlesbasedonuserinputs,fillingapsindatalineage,orfindnewchemicalcompoundssimilartoaknowncompound.  StardogprovidesanembeddedMLcapabilitythatinterfaceswith VowpalWabbit (a solutionforreinforcementlearning,supervisedlearning,andotherMLparadigms,developedbyMicrosoftResearch).ManyoftheparametersofthatlibraryareelevatedintotheStardogquery.So,youcantrainandrunamodelinasimplequery. Andlet’stouchonproductivityagain.Inaknowledgegraph,thedatacomesoutasgraphdata,andsoit’salreadysetupforanicereinforcementalgorithmtofeedthatdatabackintothegraph.Putitinaplacewhereyoucandoadditionalqueries.Youcanchecktoseeifit’svalidornot,andthatallbecomesamoreconsiderableproductivityboostfortheentireworkstream. Insummary,ifyouwanttopredictsomething,classifysomething,orseeifthingsaresimilar,Stardogcanhelpyougetstartedquickly. Infernewfacts  MLcomplementslogicalreasoning,whichtogetherprovideasuiteofreasoningcapabilitiesthatbringsforththetotalvalueofyourconnecteddata.  TraditionalAIresearchhastrendedtowardsinferencingtocapturedecision-makingandknowledgerepresentation.FromthatheritageisbornthingsliketheinferenceengineinStardog,whichisbest-in-class.Aspartofanoverallstrategy,addingintheinferencingcapabilitytoyourtoolsuite,whetheryouuseourparticularMLlibrariesoryou’redoingthatinconcertwithadditionalthird-partylibraries,you’veaddedacapabilitythatfewothershave.Andthat’stheabilitytocapturethesefactsandinfernewfacts.  Inferenceexpressesalltheimpliedandpredicatedrelationshipsandconnectionsbetweenyourdatasources,creatingaricher,moreaccurateviewofyourdata.Betterdatameansbetterlearning.Andbettermeansprovidingcontext,notjustvolume.What’sneededisAIthatcanlearnmorequicklyandproduceanswerstoquestions. Additionally,constraints ensure accurate,validdata.Use *constraints* topreventtheknowledgegraphfromaccessingbaddataortosimplyflaginconsistenciesinthedata. Integratewithexplainablemachinelearningapproaches Machineanddeeplearningsystemsareincreasinglyusedtomakedecisions.Buttherearelimitations.Evenifprovedtobehighlyaccuratedecision-makers,thesesystemscannotexplaintheirdecisionsinawaythatpeopleunderstand.There’snoexplainability.Formanyusers,thislackof “why” makesthedecisionsuntrustworthy.  Giventhatknowledgegraphsprovidecontextanddomaininformationinamachine-readableformat,youcanintegratethemwithexplainableMLapproachestoprovidemoretrustworthyexplanations. TolearnmoreaboutStardogandmachinelearning,checkoutourwhitepaper, “Machine Learning:ShiftsinBusiness.” *AscitedinGartner’s “How toBuildKnowledgeGraphsThatEnableAI-DrivenEnterpriseApplications:May27,2020 MachineLearning KnowledgeGraph KeepReading: HowtoBuildaKnowledgeGraph Feb23,2022,13minuteread Practicalstepsforbuildingknowledgegraphs:powerfultoolsforlinkeddata,dataintegration,anddatamanagement.Scaleallthoseusecasesthathavebeeninspiredbydatascience.Increaseyournumberofusers,asneeded.Andspreadtheuseofdataitself. Doyoureallyneedaknowledgegraph? Datarulestheworld.Butorganizationsstruggletoleveragethatdataforacompetitiveadvantage.Raw,uninterpreteddatainasystemsomewhereisn’tveryhelpful. UnderstandtheROIofanEnterpriseKnowledgeGraphPlatform ColleenLuther Dec8,2021,3minuteread Dataspanswideandrunsdeepacrossenterprises.It’softentuckedintopocketsandsilos.Ifdatahasn’tbeenallbroughttogetheryet,itcanbeincrediblydifficulttoparsetheexactbusinessvalueofdoingso.DatamodernizationandinnovationcanbetoughtonaildownintermsofROI. Backtoallposts downloadourfreee-guide KnowledgeGraphs101 HowtoOvercomeaMajorEnterpriseLiabilityandUnleashMassivePotential Downloadforfree Let’sstayintouch Subscribetogetourlatestcontentandstayuptodateonnewsandevents. EmailAddress Thiswebsitestorescookiesonyourcomputerwhichareusedtoimproveyourwebsiteexperienceandprovidemore customizedservicestoyou.Tofindoutmoreaboutthecookiesweuse,seeourprivacypolicy. IAccept