The Future of AI: Machine Learning and Knowledge Graphs

Graph-native learning involves computing machine learning tasks within a graph structure and takes knowledge graph augmented machine learning to the next level. Menu Close Partners FindaPartner BecomeaPartner SolutionPartners OEMPartners TechnologyPartners PartnerPortalLogin Company AboutUs Newsroom AwardsandHonors Graphs4Good Careers Culture Diversity Leadership ContactUsContactUs Support ContactUs Thankyouforyourinterest!Wewillgetbacktoyousoon! Regionalsalescontactinformation. × Products PlatformOverview→ Neo4jgraphtechnologyproductshelptheworldmakesenseof data. GraphDatabase Neo4jGraphDatabase Self-managed,deployanywhere Neo4jAuraDB Fullymanagedgraphdatabaseasaservice GraphDataScience Neo4jGraphDataScience Graphanalyticsandmodelingplatform Neo4jAuraDS New! Fullymanagedgraphdatascienceasaservice DownloadCenter Graph Database•Graph DataScience Pricing GraphDatabase•GraphData Science UserTools Neo4jDeveloperTools Desktop,Browser,andDataImporter CypherQueryLanguage Powerful,intuitive,andgraph-optimized Neo4jGraphQLLibrary Low-code,opensourceAPIlibrary Neo4jDataConnectors ApacheKafka,ApacheSpark,andBItools Neo4jBloom Easygraphvisualizationandexploration Solutions UseCases Frauddetection,knowledgegraphsandmore KnowledgeGraphs Knowledgegraphsaretheforcemultiplierofsmartdata managementandanalyticsusecases. LearnMore ByApplication AnalyticsandDataScience FraudDetection KnowledgeGraphs RealTimeRecommendations SupplyChainManagement IdentityandAccessManagement MasterDataManagement NetworkandITOperations DataPrivacy,RiskandCompliance SocialNetworking ByIndustry FinancialServices Retail LifeSciences Telecommunications Government CaseStudies In-depthlooksatcustomersuccessstories Customers Companies,governmentsandNGOsusingNeo4j ProfessionalServices Theworld’sbestgraphdatabaseconsultants Learn Resources WhoUsesNeo4j? 75percentofFortune100companies ExecutiveInsights GettoKnowGraphTechnology GraphAcademy Freeonlinecoursesandcertifications Neo4jBlog DailyreadsongeneralNeo4jtopics Videos Up-to-date,searchablearchive Books Learnindepthwithfreebooks ResourceLibrary Whitepapers,datasheetsandmore Events GraphConnect Live!June6-8,AustinTX GraphSummit LiveeventsinEMEAandAPAC Connections Ourmarqueedigitaleventseries EventsCalendar Liveonlineevents,trainingsanddemos LiveDemos WeeklydemoswithNeo4jexperts Webinars Upcomingliveandon-demandwebinars Developers DeveloperHome Bestpractices,how-toguidesandtutorials Documentation ManualsforNeo4jproducts,Cypheranddrivers DownloadCenter GetNeo4jproducts,toolsandintegrations DeveloperBlog DeepdivesintomoretechnicalNeo4jtopics Community Aglobalforumforonlinediscussion OnlineMeetups Globaldeveloperconferencesandworkshops DataScientists GraphDataScienceHome LearnwhatNeo4joffersfordatascience DataScienceDocumentation ManualfortheGraphDataSciencelibrary GraphAcademyforDataScience Freeonlinecoursesandcertificationsfordatascientists DataScienceGuides Deepdives&how-tosonmoretechnicaltopics DataScienceCommunity Aglobalforumforonlinediscussion GetStartedwithGraphDataScience DownloadoursoftwareorgetstartedinSandboxtoday! SignIn Neo4jAura Fullymanaged,cloud-nativegraphservice Neo4jSandbox Learngraphdatabasesandgraphdatascience GetStarted Neo4jAuraDB StartFree StartyourfullymanagedNeo4jclouddatabase Neo4jSandbox LearnanduseNeo4jfordatascience&more Neo4jDesktop ManagemultiplelocalorremoteNeo4jprojects Neo4jAuraDS New! Fullymanagedgraphdatascience,startingat$1/hour (Neo4jBlog)←[:BACK] TheFutureofAI:MachineLearningandKnowledgeGraphs MayaNatarajan,Sr.ProgramDirector,KnowledgeGraphs Mar11 10minsread Whileaimingtoenhancethewaypeoplesearchforinformation,knowledgegraphseasethecomplexprocessofsearchingandexplorationasalotofinformationisintheformofdata,audio,videos,andimagesaboutaperson,entity,orobject. Rightnow,nearlyhalfofDatabaseTrendsandApplicationsreadersareusingmachinelearningtobetterleveragetheirdataanddeliveranalyticalinsights. However,avarietyofchallengesexist,fromgainingaccesstotherightdatatoidentifyingoutputsthatareuseful. Theflexibilityofthegraphmodel,alongwithitsexplicitstorageofdatarelationships,makesitnotonlyeasytomanagedatacomingfromdiversesourcesbutalsotosearchandexplorethatdatatorevealnewinsightsthatwouldotherwisebeverydifficulttodiscover.Asasignofitsgrowingpopularity,inDecember2020atNeurIPS2020,oneoftheworld’slargestAIconferences,morethan136papershadtheword“graph”intheirtitles. Knowledgegraphsarearguablybothwidelyavailableandawell-understoodtechnologyforknowledgerepresentationandreasoning. BesidesreachingthepeakinGartner’shypecycleforAIin2020,knowledgegraphsareincreasinglybeingadoptedinreal-worldapplicationsbyorganizationsrangingfromindustry-leadingcorporationstomid-marketcompanies. IntroducingGraphMachineLearning Thereisalsoanotheruseofgraphsthatblossomedin2020:graphmachinelearning.Graphneuralnetworksoperateonthegraphstructures,asopposedtoothertypesofneuralnetworksthatoperateonvectors.Whatthismeansinpracticeisthattheycanleverageadditionalinformation. Graphmachinelearningalsogoesbythenameofgeometricalmachinelearning,becauseofitsabilitytolearnfromcomplexdatasuchasgraphsandmulti-dimensionalpoints. Itsapplicationsin2020havebeenrelevantinbiochemistry,drugdesign,andstructuralbiology.Knowledgegraphsandgraphmachinelearningcanworkintandem,aswell. DespitetheglobalimpactofCOVID-19,47%ofAIinvestmentswereunchangedsincethestartofthepandemicand30%oforganizationsactuallyplannedtoincreasesuchinvestments,accordingtoaGartnerpoll.Only16%hadtemporarilysuspendedAIinvestments,andjust7%haddecreasedthem. Additionalusecasesforgraphsincludethefollowing: Question-answering:Thisisoneofthemostfrequentlyusedapplicationsforknowledgegraphs.Knowledgegraphscontainawealthofinformationandquestion-answeringisagoodwaytohelpenduserstomoreeffectivelyandmoreefficientlyretrieveinformationfromknowledgegraphs. Storingresearch:Manycompanieshavebeenusingknowledgegraphsrecentlytostoreinformationgeneratedfromvariousstagesofresearchwhichcanbeusedforbuildingaccessiblemodels,riskmanagement,andprocessmonitoring. Recommendationsystems:Netflix,forexample,usesaknowledgegraphtostoreavastamountofvariedinformationforitsrecommendationsystemwhichhelpsinfindingrelationshipsbetweenmovies,TVshows,persons,etc.Later,thesepreferencesandconnectionscanbeusedtopredictwhatcustomersmightliketowatchnext. Supplychainmanagement:Companiescaneasilykeeptrackofinventoriesofdifferentcomponents,thepersonnelinvolved,andtiming,whichallowsthemtomoveitemsmoreswiftlyandcost-effectively. TheValueofCombiningKnowledgeGraphsWithAI Bringingknowledgegraphandmachinelearningtechnologytogethercanimprovetheaccuracyoftheoutcomesandaugmentthepotentialofmachinelearningapproaches.AccordingtoAndreaMalick,aresearchdirectorintheDataandAnalyticspracticeatInfo-Tech,stepsforimplementingandmaximizingthevalueofaknowledgegraphincludethefollowing: Beginwithasingleusecase,linkjustafewdatasetsandreports,andthenadddataandlinkstoitorganicallysothatit’sadynamicstructure. Onceyouhaveausecase,identifythecontentyou’llneedandclassifyitaccordingtoataxonomy.Whileyoucanrefertoindustry-standardtaxonomiesforideas,investthetimetomakethetaxonomymeaningfulforyourorganizationandunderstandhowusersorganizetheirinformation.Buyingtaxonomiesoutoftheboxorcontractingaconsultanttodoitforyouisboundtoleadtoproblems. Theorganizingstructurebecomesevenmorepowerful–anontology–whenyouusesemanticindexingtoreplaceusers’ownwordswithsynonymstobetterunderstandwhattheymean.Therequesterdoesn’tneedtoknowtheexactlabeltoretrievetheinformationtheywant. Engagebusinessusersinthecontinuousdevelopmentoftheknowledgegraph,alongwithtaxonomists,informationarchitects,anddatascientists. Adddescriptivemetadatatotheknowledgegraph,suchastheversionofthereportordatalineage,sothatuserscandecidewhetherit’stherightdataandifitsqualityisacceptable. Withknowledgegraphs,AIlanguagemodelsareabletorepresenttherelationshipsandaccuratemeaningofdatainsteadofsimplygeneratingwordsbasedonpatterns.ThisallowsAItobeamoretrustworthypartneraswesearchtheweb.–StephanieSimone TheConvergenceofAIandKnowledgeGraphs CompaniesareincreasinglyusingAIapplicationsfordecisionmaking.However,duetolackofcontextualinformation,AIsystemshavenotyetbeenabletoachievetheirfullpotentialasreliablesolutionsforcomplexproblems. Enterknowledgegraphs:alogicalwaytocapturedatarelationshipsandconveytheirmeaning.KnowledgegraphsdriveintelligenceintothedataitselfandgiveAIthecontextitneedstobemoreexplainable,accurate,andrepeatable. NeitherAInorknowledgegraphsisnewtechnology,butonlylatelyhavetheycomeofageandjoinedforces.Althoughdataandcomputepowerhavecontributedtotheirriseinthelastdecade,it’sthepowerfulcombinationofthetwothatisspurringanexplosionofinterestinContextualAI. DrivingIntelligenceIntoDataUsingKnowledgeGraphs Knowledgeiseverycompany’smostprizedasset;however,itsvalueislimitedunlessorganizationscanleveragethatknowledgeinthecorrectcontext.Thisiswhereknowledgegraphscomeintoplay. Knowledgegraphsexplicitlysurfacetherichrelationshipsbetweendatathatexperienceddomainexpertsnaturallyconsider.Afterall,inreality,therearenoisolatedpiecesofdata,butonlyrich,connecteddomainsallaroundus.Aknowledgegraphplacesdataincontextbyestablishingconnectionsamongdata. Aknowledgegraphthenenrichesthedata’smeaningandutilitybyaddingalayerofsemantics,therebyallowingsoftwareagentstoreasonaboutit.Byaddingrelationshipstodataandenhancingitwithsemantics,knowledgegraphsdriveintelligenceintodata,makingitsmarter. KnowledgeGraphsEnhanceMLFromSourcingtoTrainingtoPredictions AIandmachinelearningareplayinganever-increasingroleinenterprisestoday.Machinelearningisusedineveryindustry:inhealthcaretodetectcanceroustumors,insupplychainstofindfactorsthatpositivelyandnegativelyimpactbusiness,andinfinancialservicestoallowinvestorstoidentifynewopportunitiesorknowwhentotrade. Interestingly,machinelearningisenhancedusingknowledgegraphsbecauseoftheirinnateabilitytosurfacecontext.Contextualinformationisknowntoincreasepredictiveaccuracy,makedecisioningsystemsmoreflexible,andprovideaframeworkfortrackingdatalineage. Machinelearningdependsondata;themorethedata,thehigherthedataquality,andthemoredatavariety,thebettertheresults.Unfortunately,mostdatascienceapproachesleaveoutcontextualinformationbecauseconnectionsanddatastructuresaredifficulttoprocess.However,knowledgegraphscapture,persist,andmakethiscontextualinformationusable,whichmeanstheycanenhanceeverystepofthemachinelearningprocess.Fromdatasourcingandtrainingmachinelearningmodelstoanalyzingpredictionsandapplyingresults,contextualizedAIsystemsaremorereliable,robust,explainable,andtrustworthy. Intheinitialstepofdatasourcing,knowledgegraphsareusedfordatalineagetotrackthedatathatfeedsmachinelearning–wherethedatacamefrom,howthedatachanged,wherethedataisused,andwhousedit.Ifyoucan’ttrustthedatausedforML,youcan’ttrusttheresults.Datalineageandmasterdatamanagementusingknowledgegraphsalsoserveasanaudittrailforcompliance,especiallyinregulatedindustries. Thenextphaseinmachinelearningistrainingamachinelearningmodel,whichinvolvesprovidinganMLalgorithmwithtrainingdataandsignificantfeaturestolearnafunctionformakingpredictions.Machinelearningmodelswithoutcontextrequireexhaustivetraining,strictlyprescriptiverules,andcanonlybeappliedtospecificapplications.Knowledgegraphsaddthemuch-neededcontextthatresultsinbetterpredictions–allwithexistingdata.Knowledgegraphsalsoallowforgraphfeatureengineeringusingsimplegraphqueriesand/ormorecomplexgraphalgorithms.Weknowthatrelationshipsarehighlypredictiveofbehavior,sousingtheseconnected,contextualfeaturesmaximizesthepredictivepowerofmodelswhileincreasinghowbroadlyasolutioncanbeapplied. Onceamachinelearningmodelhasbeendeveloped,itisessentialtounderstandifthemodelisusefulandifthemodelismakingcorrectpredictions.Knowledgegraphswithincorporatedrelationshipinformationallowforeasygraphinvestigationsandcounterfactualanalysisbydomainexperts.Anexpertmighttesthypothesesbyexploringsimilarcommunitiesintheknowledgegraphordebugoddresultsbydrillingintohierarchiesanddependencies. Knowledgegraphsbuiltongraphtechnologieshavesignificantadvantagesasgraphsnaturallystore,compute,andanalyzeconnectionsandrelationshipsamongdata.Moreover,graphalgorithmsarespecificallydevelopedtoleveragethetopologyofdatathroughconnections:findcommunities,uncoverinfluentialcomponents,andinferpatternsandstructure.Incorporatingthepredictiveelementsofcontextfromaknowledgegraphintomachinelearningnotonlyincreasesaccuracybutreducesfalsepositives. Graph-nativelearninginvolvescomputingmachinelearningtaskswithinagraphstructureandtakesknowledgegraphaugmentedmachinelearningtothenextlevel.Itprovidestheabilitytolearngeneralized,predictivefeaturesdirectlyfromwithinthegraphwithoutknowingwhatdatastructuresaremostpredictive.Thisissignificantasorganizationsdon’talwaysknowwhichfeaturesaremostimportant,letalonehowtorepresentconnecteddataforuseinmachinelearningmodels. Theconvergenceofmachinelearningandknowledgegraphsisjustthebeginning.Asindustriesbegintoseethepowerincombiningthesetechnologies,expecttoseeincreasingdemandforintegratedsolutionsandstreamlinedworkflows. LeveragingKnowledgeGraphsandMLforCompetitiveAdvantage Companiestodayareleveragingknowledgegraphswithmachinelearningformanyusecases,frommerelyenhancingheuristicstomorecomplexusesliketrainingembeddingsinagraph-nativelearningmodel.Examplesinclude: Aglobale-commerceleaderthathascreatedashoppingbotusingaknowledgegraphtoaddcontexttomachinelearningtomakebetterheuristicdecisionsaboutuserintent. AFortune100construction-equipmentmanufacturerthatsuccessfullyutilizesmachinelearningandaknowledgegraphforpredictivemaintenanceandimprovingequipmentlifespan.Inthiscase,millionsofwarrantyandservicedocumentsareparsedfortextandaddedtoaknowledgegraphforcontextsothatmachinelearningmodelscanlearn“primeexamples”andanticipaterequiredmaintenance.Thispredictivemaintenancehasallowedthecompanytoproactivelytakeremedialaction,savedowntimeandassociatedcostsandincreaseproductivity. Aglobalpharmaceuticalcompanythatiscombiningaknowledgegraph,graphqueriesandgraphalgorithmswithtraditionalmachinelearningapproachestomapandpredictpatientjourneys.Billionsofrecordsoverathree-yearperiodarebeingusedtoextractpatientpathsandtrainembeddingsforpredictingsuccessfulinterventionpointsandimprovingpatientoutcomes. Neo4jKnowledgeGraphsinAction Knowledgegraphslendthemselvesnaturallytomanyusecases.Themajorityoftheseusecasesfallalongaspectrumofthreemajorcategories:datamanagement,datadiscovery,anddataanalytics. Datamanagementusecasestendtobethemostdistinctandfocusondataaggregation,validation,andgovernance.Dataassuranceisthemotivationoftheseusecases,includingdatacatalogs,datalineageandprovenance,compliance,datagovernance,riskmanagement,anddatafabric.Dataassuranceknowledgegraphsprovideincreasedtrustandexplainability. Datadiscoveryusesgobeyondsimpleinformationvisibility,emphasizinguser-ledexploration,deduction,andinferenceofnewknowledge.Theimpetusformanyoftheseusecasescentersaroundexpertunderstandingandinsightswithexamplessuchascustomer,patient,orproduct360,sharinglessonslearned,investigations(criminal,anti-moneylaundering,fraud),andrecommendations.Insightknowledgegraphsdelivercompletevisibilityandimprovedproductivity. Dataanalyticsanddatadiscoveryusecasesoverlap,butthegoalistoimproveforecastsandpredictionsandprescribeoptimalactions.Althoughunderstandingisstillessential,theultimategoalhereisbetterdecisioning.Afewexamplesarecustomerorpatientjourneyanalysis,what-ifanalysis,nextbestactionoroptions,churnanalysis,featureengineeringformachinelearning,andfindingcommunitiesandinfluencersincomplexsystemsforinvestigations(criminal,anti-moneylaundering,fraud)andrecommendations.Decisioningknowledgegraphsrenderbetterpredictionsandmorebreakthroughsthroughthepowerofanalytics. TheBottomLine BothAIandknowledgegraphsaredrivingthenextwaveofcompetitiveadvantageforcompanies.Butthequestioncomesdowntoexecutionandwhichcompaniescanusethemtogethersuccessfully–whethertoreducetheriskoffraud,improvepatientoutcomes,makebetterinvestmentdecisions,orincreaseemployeeproductivity.Areyouready? Readytodivedeeperintotheworldofknowledgegraphs?GetthefreewhitepaperFromGraphtoKnowledgeGraph:AShortJourneytoUnlimitedInsights. GetMyFreeCopy Emailmeblogupdates! Theinformationyouprovidewillbeusedinaccordancewiththetermsofourprivacypolicy. aiAI/MLartificialintelligencecontextContextualAIdatasourcingGraphMachineLearningknowledgegraphsmachinelearningML Neo4jCommunity Disclaimer Author MayaNatarajan,Sr.ProgramDirector,KnowledgeGraphs MayaNatarajanisresponsibleforthego-to-marketstrategyforknowledgegraphsatNeo4j.Sheispassionateaboutbringingdifferenttechnologiestogethertosolvecomplexproblemsandischampioningtheuseofknowledgegraphstobringcontexttovarioussystems.MayahaspositionedtechnologiesfromBlockchaintoPredictive&User-BasedAnalytics...knowmore RelatedArticlesGraphThatSavesLives:ConnectingResearchPaperstoClinicalTrialstoSavingHumanLivesJul076minsreadGraphDataScienceonGoogleCloud:Neo4jAuraDSandVertexAIJun147minsreadThisWeekinNeo4j:GraphConnect,NewToolsandFeatures,Neo4jGDSPythonClient,ImageToGraph,andMoreJun112minsread