Graph Exploration: Taking the User into the Loop Exploration: Taking the User into the Loop...

Graph Exploration: Taking the User into the Loop

Davide Mottin,AnjaJentzsch,EmmanuelMüllerHasso Plattner Institute,Potsdam,Germany

2016/10/24CIKM2016,Indianapolis,US

CIKM 2016 TUTORIAL

Where we are

D. MOTTIN, A. JENTZSCH, E. MÜLLER

ExploratoryGraphAnalysis(35min)

RefinementofQueryResults(35min)

FocusedGraphMining(35min)

103

Background(5 min)Graphmodels,subgraphisomorphism,subgraphmining,graphclustering

RealWorld-UseCase(15min)LinkedDatagraphs

Challengesanddiscussion

CIKM 2016 TUTORIAL

The Web of Data

• 1,019datasets• 84+billionRDFtriples• 808+millionRDFlinksbetweendatasets

D. MOTTIN, A. JENTZSCH, E. MÜLLER 104

http://lod-cloud.net

CIKM 2016 TUTORIAL

Vocabularies on the Web of Data• TheWebofDataisheterogeneous⁃ Manyvocabulariesareinuse(576asofOctober2016)⁃ Manydifferentwaystorepresentthesameinformation

D. MOTTIN, A. JENTZSCH, E. MÜLLER 105

https://lov.okfn.org

CIKM 2016 TUTORIAL

RDF Data Model

D. MOTTIN, A. JENTZSCH, E. MÜLLER 106

ns:cikm2016 ns:Event

dbpedia:Indianapolis

ACMConferenceonInformationandKnowledgeManagement(CIKM2016)

rdf:type

rdfs:label

ns:location

CIKM 2016 TUTORIAL

RDF Data Model

D. MOTTIN, A. JENTZSCH, E. MÜLLER 107

ns:cikm2016 ns:Event

dbpedia:Indianapolis

ACMConferenceonInformationandKnowledgeManagement(CIKM2016)

rdf:type

rdfs:label

ns:location

820445

dbpedia:populationTotal

JosephH.Hogsett

dbpedia:leaderName

CIKM 2016 TUTORIAL

Linked Data exploration use cases

• Datasetexploration• Graphmining• Queryformulationandrefinement

! ButLinkedDataismessy

D. MOTTIN, A. JENTZSCH, E. MÜLLER 108

CIKM 2016 TUTORIAL

Linked Data graph exploration challenges

● Nestedgraphs � Makesreasoningdifficult

● Loosestructure � Thingshavedifferentpropertysets

● Incomplete � Missingpropertydefinitions

● Poorlyformatted � Propertytypesusedinconsistently

● Inconsistent � Multiplerepresentationsclaimoppositethings

D. MOTTIN, A. JENTZSCH, E. MÜLLER 109

CIKM 2016 TUTORIAL

Linked Data exploration systems timeline

D. MOTTIN, A. JENTZSCH, E. MÜLLER 110

CIKM 2016 TUTORIAL

DBpedia Mobile

D. MOTTIN, A. JENTZSCH, E. MÜLLER 111

C.BeckerandC.Bizer.DBpedia mobile:Alocation-enabledlinkeddatabrowser.LDOW2008.

• displaysWikipediadataonmap• aggregatesdifferentdatasources

CIKM 2016 TUTORIAL

RelFinder

D. MOTTIN, A. JENTZSCH, E. MÜLLER 112

Heim,P.,Hellmann,S.,Lehmann,J.,Lohmann,S.,andStegemann,T.RelFinder:RevealingRelationshipsinRDFKnowledgeBases.SAMT2009.

• visualizationofpathsbetweenany2entities• pathidentificationoninstancelevel

CIKM 2016 TUTORIAL

gFacet

D. MOTTIN, A. JENTZSCH, E. MÜLLER 113

P.Heim,T.Ertl,andJ.Ziegler.Facetgraphs:Complexsemanticqueryingmadeeasy.TheSemanticWeb:ResearchandApplications.Springer,2010.

• Schemaexploration

• combinesgraph-basedvisualizationand

facetedfilteringtechniques

CIKM 2016 TUTORIAL

graphVizdb

D. MOTTIN, A. JENTZSCH, E. MÜLLER 114

Bikakis,N.,Liagouris,J.,Krommyda,M.,Papastefanatos,G.andSellis,T.graphVizdb:Ascalableplatformforinteractivelargegraphvisualization.ICDE,2016

• Graphlayoutisindexedwithaspatialdatastructure,i.e.,anR-tree,andstoredinadatabase

• Inruntime,useroperationsaretranslatedintoefficientspatialoperations(i.e.,windowqueries)inthebackend

CIKM 2016 TUTORIAL

LODeX

D. MOTTIN, A. JENTZSCH, E. MÜLLER 115

Benedetti,F.,Bergamaschi,S.andPo,L.Lodex:Atoolforvisualqueryinglinkedopendata.ISWC,2015

• ExploreaLinkedDatasetusingaschemasummary

• Pickgraphicalelementsfromittocreateavisualquery

• Browsetheresults• Refinethequery

CIKM 2016 TUTORIAL

Aemoo

D. MOTTIN, A. JENTZSCH, E. MÜLLER 116

A.Musetti,A.G.Nuzzolese,F.Draicchio,V.Presutti,E.Blomqvist,A.Gangemi,andP.Ciancarini.Aemoo:Exploratorysearchbasedonknowledgepatternsoverthesemanticweb.SemanticWebChallenge,2012.

• ExploratorysearchsystembasedonEncyclopedicKnowledgePatterns

• EKPareknowledgepatternsthatdefinethetypicalclassesusedtodescribeentitiesofacertainclass

CIKM 2016 TUTORIAL

Linked Jazz

D. MOTTIN, A. JENTZSCH, E. MÜLLER 117

M.C.Pattuelli,M.Miller,L.Lange,S.Fitzell,andC.Li-Madeo.Craftinglinkedopendataforculturalheritage:Mappingandcuration toolsforthelinkedjazzproject.Code4LibJournal,2013.

• revealsthenetworkofthesocialandprofessionalrelationswithintheAmericanjazzcommunity

CIKM 2016 TUTORIAL

Semantic Wonder Cloud

D. MOTTIN, A. JENTZSCH, E. MÜLLER 118

http://sisinflab.poliba.it/semantic-wonder-cloud/index/

CIKM 2016 TUTORIAL

inWalk

D. MOTTIN, A. JENTZSCH, E. MÜLLER 119

Castano,S.,Ferrara,A.andMontanelli,S.inWalk:InteractiveandThematicWalksinsidetheWebofData.EDBT,2014

CIKM 2016 TUTORIAL

ProLOD++ Mining Graph Patterns on the Web of Data

D. MOTTIN, A. JENTZSCH, E. MÜLLER 120

Jentzsch,A.,Dullweber,C.,Troiano,P.,Naumann,F.ExploringLinkedDataGraphStructures.ISWC2015.

ProLOD++● WebframeworkforvariousdataprofilingandminingtasksonLinkedDatasets

● ExplorativeresearchonLinkedDatasetgraphstofind⁃ frequentgraphpatterns⁃ commongraphpatternsforclasses⁃ generalgraphmodelforLinkedDatasets

https://prolod.org

CIKM 2016 TUTORIAL

ProLOD++Graph pattern mining

D. MOTTIN, A. JENTZSCH, E. MÜLLER 121

Jentzsch,A.,Dullweber,C.,Troiano,P.,Naumann,F.ExploringLinkedDataGraphStructures.ISWC2015.

DefinitionofcoresetoffrequentgraphpatternsinLinkedDatasetsbasedonsatellitecomponentanalysis

CIKM 2016 TUTORIAL

ProLOD++Graph patterns

D. MOTTIN, A. JENTZSCH, E. MÜLLER 122

Jentzsch,A.,Dullweber,C.,Troiano,P.,Naumann,F.ExploringLinkedDataGraphStructures.ISWC2015.

● Groupclass-coloured graphsbytheirpermutationgroups[Luks82]⁃ Permutationgroup:thesetofallautomorphisms ofagraph

drugtargetunknown

CIKM 2016 TUTORIAL

Loupe

D. MOTTIN, A. JENTZSCH, E. MÜLLER 123

Mihindukulasooriya,N.,Poveda-Villalón,M.,García-Castro,R.and Gómez-Pérez,A.Loupe - An OnlineTool for InspectingDatasets inthe Linked DataCloud.ISWC2015.

• Frequenttriplepatterns• Graphicalontologybrowsing

CIKM 2016 TUTORIAL

Requirements for Linked Data exploratory search systems

D. MOTTIN, A. JENTZSCH, E. MÜLLER 124

• Thesystemprovidesefficientoverviews• Thesystemhelpstheusertounderstandtheinformationspaceandtoshapehismentalmodel

• Theusercanexploremultiple,heterogeneousresultsandbrowsingpaths

• Thesystemeasesthememorizationofrelevantresults

• Thesysteminspirestheuserandshapeshisinformationneed• Thesystemprovokesdiscoveries

CIKM 2016 TUTORIAL

Challenges

D. MOTTIN, A. JENTZSCH, E. MÜLLER 125

● Displayingthegraphforexploration⁃ E.g.byclusteringoftopicaldomains⁃ Allowingtheusertodrilldown

● Livegraphexploration⁃ E.g.viafederatedSPARQLqueries▪ RequiresknowledgeonendpointURIs▪ Slowinreal-time

● Guidingtheusertointerestingpartsofthegraph⁃ Usuallydonebyentityinlinks▪ Limitedinsights

CIKM 2016 TUTORIAL

Tutorial outline

D. MOTTIN, A. JENTZSCH, E. MÜLLER

ExploratoryGraphAnalysis(35min)

RefinementofQueryResults(35min)

FocusedGraphMining(35min)

126

Background(5 min)Graphmodels,subgraphisomorphism,subgraphmining,graphclustering

RealWorld-UseCase(15min)LinkedDatagraphs

Challengesanddiscussion

CIKM 2016 TUTORIAL

Summary of Exploratory Graph Analysis

D. MOTTIN, A. JENTZSCH, E. MÜLLER 127

ApproximateQueries• UserqueryisimpreciseBy-Examplemethods• Userqueryisanexampleresult

• Onlyneedapartialknowledgeonthedata

• Noneedforcomplicatequerylanguages(useexamples,partialdescriptions)

• Thequeryadaptstouserneed• Enableexploratorysearchby

usingsmallqueriesonthedata

Query(agraph)

Graph

?

Query(anexample)

Graph

CIKM 2016 TUTORIAL

Challenges for Exploratory Graph Analysis

D. MOTTIN, A. JENTZSCH, E. MÜLLER 128

• Unsupportedinmostofthecurrentgraphdatabases• No”universal”indextoanswermultipletypeofqueries• Partitioningonlyforexactqueryanswering

Database

• Userinteractivityintheexplorationprocess• Nosolutionsforprobabilisticgraphs• Respondtoquerieswhilethegraphchanges• Findexamplesinstreamingsettings

InformationretrievalInformation

retrieval

• Exploitingquerylogsforpersonalizedqueryanswering• Retrieveresultsinformofdocumentsconvertingthequery

structures

Datamining

CIKM 2016 TUTORIAL

Summary of Focused Graph Mining

D. MOTTIN, A. JENTZSCH, E. MÜLLER 129

Thefocusonindividualuserinterest…asQuery totheGraphMiningSystem…asSeedNode(s) forLocalSearch…asAttributes andWeights

• getorinferuserinterestà unexpectedresults

• interactiveexplorationà intuitiveparametrization

• adaptivegraphminingà individuallocalsearch

CIKM 2016 TUTORIAL

Challenges for Focused Graph Mining

D. MOTTIN, A. JENTZSCH, E. MÜLLER 130

Datamining

scale

Userinteractivityinthegraphminingprocess• unsupportedinmostofthecurrentgraphminingalgorithms• hugevarietyofuserinteractions possible• feedbackloopneedstobeunified andbecomeexchangeable

Revolutionofformalmodelsandsearchalgorithms• insufficientextensionsofexistingmodelsandalgorithms• adaptivesteering ofalgorithmsvs.fixedparametrization• evaluationofalgorithmswithuserstudies

Scalabilityofalgorithmsforreal-timeinteraction• NP-hardproblems,heuristicalgorithms,…,stillnotscalable• exploittheuserinterest forpruningthesearchspace

CIKM 2016 TUTORIAL

Summary of Refinement of Query Results

D. MOTTIN, A. JENTZSCH, E. MÜLLER 131

Refinement• Theuserqueryistoorestrictiveortoo

genericTop-kResults• QueriestypicallyhaveinexactmatchesSkylineQueries• Findsmallsetofinterestingitemswith

manydimensionsandincrementalupdates

• Theusermighthaveaverygenericideaofhowtodescribethestructureofinterest

• Thesystemguidestheusertowardstheanswerwithsimplesteps

• Theresultsareexplainedwithreformulations

• Thequerymatchesareinexactandinteresting

CIKM 2016 TUTORIAL

Challenges for Refinement of Query Results

D. MOTTIN, A. JENTZSCH, E. MÜLLER 132

• Realtimeperformance• Profilingofqueriesforoptimizedperformance

• Personalizedreformulationsandinteractivity• Facetsearchdiscoveryingraphs

• Uncertaingraphdata

Database

InformationretrievalInformation

retrieval

Datamining

CIKM 2016 TUTORIAL

The missing tiles in graph exploration

D. MOTTIN, A. JENTZSCH, E. MÜLLER 133

Interactivity Adaptivity

Personalization Scalability

Questions?

CIKM 2016 TUTORIAL D. MOTTIN, A. JENTZSCH, E. MÜLLER 134

Slides:https://hpi.de//mueller/tutorials/graph-exploration.html