Ontology-Aware Partitioning for Knowledge Graph Identification Jay - - PowerPoint PPT Presentation

SLIDE 1

Ontology-Aware Partitioning for Knowledge Graph Identification

Jay Pujara, Hui Miao, Lise Getoor, William Cohen Workshop on Automatic Knowledge Base Construction 10/27/2013

SLIDE 2

Knowledge Graph Ingredients: Entities

Baltimore New York San Francisco New York California Maryland Baseball Football Giants Giants Orioles

SLIDE 3

Knowledge Graph Ingredients: Labels

Baltimore New York San Francisco New York California Maryland Baseball Football Giants Giants Orioles

City Location State Sport SportsTeam

SLIDE 4

Knowledge Graph Ingredients: Relations

Baltimore New York San Francisco New York California Maryland Baseball Football Giants Giants Orioles

City Location State Sport SportsTeam

SLIDE 5

Knowledge Graph Nuisance: Noise!

Baltimore New York San Francisco New York California Maryland Baseball Football Giants Giants Orioles

City Location State Sport SportsTeam

SLIDE 6

Knowledge Graph Ingredients: Ontology

Baltimore New York San Francisco New York California Maryland Baseball Football Giants Giants Orioles

City Location State Sport SportsTeam

SLIDE 7

Ontological rules for Knowledge Graphs

Adapted from Jiang et al., ICDM 2012

Inverse: wO : Inv(R, S) ˜ ∧ Rel(E1, E2, R) ⇒ Rel(E2, E1, S) Selectional Preference: wO : Dom(R, L) ˜ ∧ Rel(E1, E2, R) ⇒ Lbl(E1, L) wO : Rng(R, L) ˜ ∧ Rel(E1, E2, R) ⇒ Lbl(E2, L) Subsumption: wO : Sub(L, P) ˜ ∧ Lbl(E, L) ⇒ Lbl(E, P) wO : RSub(R, S) ˜ ∧ Rel(E1, E2, R) ⇒ Rel(E1, E2, S) Mutual Exclusion: wO : Mut(L1, L2) ˜ ∧ Lbl(E, L1) ⇒ ˜ ¬Lbl(E, L2) wO : RMut(R, S) ˜ ∧ Rel(E1, E2, R) ⇒ ˜ ¬Rel(E1, E2, S)

SLIDE 8

Knowledge Graph Identification (KGI)

• Joint inference over possible knowledge graphs
• Resolves co-referent entities
• Removes spurious labels and relations
• Infers missing labels and relations
• Uses many uncertain sources
• Enforces ontological constraints

Pujara, Miao, Getoor, Cohen, "Knowledge Graph Identification" International Semantic Web Conference, 2013

SLIDE 9

KGI: Under the Hood

• Define a probability distribution over knowledge graphs:
• Hinge-Loss Markov Random Fields
• Templated: easy to define with probabilistic soft logic
• Continuous: atoms have continuous truth values in [0,1] range
• Efficient: inference via convex optimization in O(|R|)
• Superior speed and quality for KGI (Pujara, ISWC13)

P(G | D) = 1 Z exp − wr

r∈R

∑

ϕr(G) $ % & '

Pujara, Miao, Getoor, Cohen, "Knowledge Graph Identification" International Semantic Web Conference, 2013

Grounding of logical rule in model weighted potential takes the form of a hinge-loss

SLIDE 10

Problem: Knowledge Graphs are HUGE

academicprogramatuniversity academicfield acquired company acquiredby actor agentactsinlocation agent agentbelongstoorganization humanagent agentcausedevent agentcompeteswithagent agentcontributedtocreativework agentcontrols agentcreated agenthaswebsite agenthierarchicallyaboveagent agenthierarchicallybelowagent agentinteractswithagent agentinvolvedwithitem agentleadsorganization agentowns agentparticipatedinevent agentrepresentsorganization agentstudiesphysiologicalcondition person ageofperson nonneginteger agriculturalproductproducedbyfarm agriculturalproduct airportincity airport animaldevelopdisease animal animaleatfood animaleatvegetable animalistypeofanimal animalpredators animalpreyson animaltypehasanimal aquariumincity aquarium atdate everypromotedthing athletealsoknownas athlete athletehomestadium athleteplaysforteam athleteplaysinleague athleteplayssport athleteplayssportsteamposition atlocation attractionofcity attraction automakerproducesmodel automobilemaker automodelproducedbymaker automobilemodel sportsteam bakedgoodservedwithbeverage bakedgood biologicalfatherofperson bodypart book brotherof buildinglocatedincity building ceoof ceo chemicalistypeofchemical chemical cityalsoknownas city cityaquariums cityattractions citycapitalofcountry citycapitalofstate citycontainsbuilding cityhasairport cityhascompanyoffice cityhashospital cityhasshoppingmall cityhasstreet cityhastrainstation cityhastransportation cityhotels citylanguage cityleadbyperson cityliesonriver citylocatedincountry citylocatedingeopoliticallocation citylocatedinstate citymuseums citynewspaper cityofpersonbirth cityparks cityradiostation agentrelatedtolocation athletecoach chemicaltypehaschemical cityrestaurants countrycities country drughassideeffect drug inverseofemotionassocietedwithdisease itemexistsatdate item leaguestadiums sportsleague movie

• rganizationhiredperson
• rganization

personhascitizenship physiologicalconditionstudedbyperson physiologicalcondition politicianusholdsoffice schoolattendedbyperson school sportsportsgameexample sport statelocatedincountry stateorprovince teamwontrophy citysportsteams citystadiums citytelevisionstation cityuniversities cityzoos clothingtogowithclothing clothing coach coachesathlete coachesinleague coachesteam coachwontrophy companiesheadquarteredhere companyalsoknownas companyceo companyeconomicsector competeswith controlledbyagent countrycapital drugpossiblytreatsphysiologicalcondition economicsectorcompany economicsector emotionassociatedwithdisease emotion equipmentusedbysport sportsequipment eventatlocation event eventcausedbyagent eventcausedbyevent eventcausedevent eventhasparticipantagent eventinvolveditem farmlocatedincountry farm farmlocatedinstate farmproducesagriculturalproduct fatherofperson foodcancausedisease food fooddecreasestheriskofdisease furniturefoundinroom furniture geopoliticallocationcontainscity geopoliticallocation geopoliticallocationcontainscountry geopoliticallocationcontainsstate geopoliticallocationresidenceofpersion geopoliticalorganizationleadbyperson geopoliticalorganization hasbrother hasfamilymember hasofficeincity hasofficeincountry hassibling headquarteredin hospitalincity hospital hotelincity hotel instrumentplayedbymusician musicinstrument inverseofanimaldevelopdisease inverseofanimaleatfood inverseofanimaleatvegetable vegetable inverseofbakedgoodservedwithbeverage beverage inverseofclothingtogowithclothing inverseofdateofevent countrycontainsfarm countrycurrency countryhascitizen countryhascompanyoffice countrylanguage countryleadbyperson countrylocatedingeopoliticallocation countryproducesproduct countrystates creativeworkcontributedtobyagent creativework currencycountry currency date dateof dateofevent dateofpersondeath deathageofperson director inverseoffoodcancausedisease inverseoffooddecreasestheriskofdisease inverseofinvertebratefeedonfood inverseofplantgrowinginplant plant inverseofplantrepresentemotion inverseofriveremptiesintoriver river inverseofvegetableproductioninstateorprovince inverseofweaponmadeincountry inverseofwriterwasbornincity invertebrate issueofpoliticsbill politicsissue issueofpoliticsgroup itemexistsatlocation itemfoundinroom iteminvolvedwithagent itemusedatlocation itemusedwithitem journalist lakeinstate lake languageofcity language languageofcountry languageofuniversity latitudelongitude latitudelongitudeof llcoordinate leaderofcountry leaguecoaches leagueplayers leagueteams locatedat location locationactedinbyagent locationactedinbyorganization locationcontainslocation locationexistsatdate locationlocatedwithinlocation locationofevent locationofitemexistence locationofitemuse locationofpersonbirth locationrelatedtoagent locationrepresentedbypolitician locationresidenceofperson losingscoreofsportsgame marriedinyear mlareaexpert mlarea mlauthorexpertin mlauthor mountaininstate mountain museumincity museum musicartistgenre musicartist musicartistmusician musicgenreartist musicgenre musicianinmusicartist musician musicianplaysinstrument newspaperincity newspaper

• fficeheldbypolitician

politicaloffice

• fficeheldbypoliticianus
• rganizationacronymhasname
• rganizationactsinlocation
• rganizationhasagent
• rganizationhasperson
• rganizationleadbyagent
• rganizationleadbyperson
• rganizationnamehasacronym
• rganizationrepresentedbyagent
• rganizationterminatedperson
• wnedbyagent

parentofperson parkincity park personattendsschool personbelongstoorganization personbirthdate personbornincity personborninlocation persondiedatage persongraduatedschool personhasage personhasfather personhasjobposition personhasparent personhasresidenceingeopoliticallocation personhasresidenceinlocation personhiredbyorganization politician personleadsgeopoliticalorganization personleadsorganization personterminatedbyorganization personwrittenaboutinpublication physiologicalconditionpossiblytreatedbydrug politicsbillconcernsissue politicsbill politicsbillsponsoredbypolitician politicsbillsponsoredbypoliticianus politicsgroupconcernsissue politicsgroup producedby product producesproduct producesproducttype productinstances productproducedincountry professionistypeofprofession profession professiontypehasprofession professionusestool proxyfor proxyof publication publicationwritesabout radiostationincity radiostation relatedto religionusesplaceofworship religion restaurantincity restaurant riveremptiesintoriver riverflowsthroughcity roomcancontainfurniture room roomcancontainitem schoolgraduatedperson scoreofsportsgame gamescore shoppingmallincity shoppingmall sideeffectcausedbydrug sporthassportsteamposition sportsgame sportsgameloserscore sportsgamescore sportsgamesport sportsgamewinnerscore sportsteampositionathlete sportsteamposition sportsteampositionforsport sportteam sportusesequipment sportusesstadium stadiumalsoknownas stadiumoreventvenue stadiumhometeam stadiumhometoathlete stadiumhometoleague stadiumhometosport stadiumlocatedincity statecontainsfarm statehascapital statehaslake statehasmountain placeofworshippracticesreligion placeofworship plantgrowinginplant plantincludeplant plantrepresentemotion players politicianholdsoffice politiciansponsoredpoliticsbill statelocatedingeopoliticallocation streetincity street subpartoforganization superpartoforganization teamalsoknownas teamcoach teamhomestadium teammember teamplaysagainstteam teamplaysincity teamplaysinleague teamplayssport televisioncompanyaffiliate televisionnetwork televisionstationaffiliatedwith televisionstation toolusedbyprofession tool trainstationincity trainstation transportationincity transportation trophywonbycoaches awardtrophytournament trophywonbyteam typeproducedby university vegetableproductioninstateorprovince vehicle visualartist visualartistartmovement visualartmovementartist visualartmovement weaponmadeincountry weapon websiteoperatedbyagent website wineproducedbywinery wine wineryproduceswine winery winningscoreofsportsgame worker worksfor writerwasbornincity writer zooincity zoo abstractthing beach cave continent county highway island landscapefeatures mountainrange planet trail buildingmaterial celltype charactertrait cognitiveactions color ethnicgroup game geometricshape geopoliticalentity hobby medicalprocedure mlalgorithm mldataset mlmetric perceptionaction perceptionevent physicalaction physicalcharacteristic physicsterm programminglanguage protein recipe researchproject sociopolitical species architect astronaut chef criminal judge monarch scientist buildingfeature consumerelectronicitem fungus householditem mediatype

• fficeitem

candy cheese condiment bridge monument retailstore skyscraper zipcode amphibian archaea bacteria bird fish mammal reptile nonprofitorganization port professionalorganization skiarea terroristorganization tradeunion arachnid crustacean insect mollusk vertebrate comedian model artery bone braintissue lymphnode muscle nerve vein arthropod professor automobileengine personalcareitem software bank biotechcompany creditunion mediacompany fruit meat bathroomitem bedroomitem hallwayitem kitchenitem conference convention election eventoutcome filmfestival militaryeventtype musicfestival weatherphenomenon governmentorganization nongovorganization grain grandprix

• lympics

race highschool lyrics musicalbum musicsong poem televisionshow sportsevent blog magazine recordlabel nondiseasecondition nut parlourgame boardgame cardgame videogame personafrica personantarctica personasia personaustralia personeurope personnorthamerica personsouthamerica personbylocation personcanada personmexico traditionalgame celebrity

SLIDE 11

Problem: Knowledge Graphs are HUGE

SLIDE 12

Solution: Partition the Knowledge Graph

SLIDE 13

Partitioning: advantages and drawbacks

• Advantages
• Smaller problems
• Parallel Inference
• Speed / Quality Tradeoff
• Drawbacks
• Partitioning large graph time-consuming
• Key dependencies may be lost
• New facts require re-partitioning

SLIDE 14

Key idea: Ontology-aware partitioning

• Partition the ontology graph, not the knowledge graph
• Induce a partitioning of the knowledge graph based on the
• ntology partition

City State Location SportsTeam Sport

citySportsTeam teamPlaysInCity teamPlaysSport Mut D

• m

R n g Dom Rng Inv locatedIn Rng

SLIDE 15

Considerations: Ontology-aware Partitions

• Advantages:
• Ontology is a smaller graph
• Ontology coupled with dependencies
• New facts can reuse partitions
• Disadvantages:
• Insensitive to data distribution
• All dependencies treated equally

SLIDE 16

Refinement: include data frequency

• Annotate each ontological element with its frequency
• Partition ontology with constraint of equal vertex weights

City State Location SportsTeam Sport

citySportsTeam teamPlaysInCity teamPlaysSport Mut D

• m

R n g Dom Rng Inv locatedIn Rng 2719 1171 1706 822 15391 7349 1177 10 2568

SLIDE 17

Refinement: weight edges by type

• Weight edges by their ontological importance

City State Location SportsTeam Sport

citySportsTeam teamPlaysInCity teamPlaysSport Mut D

• m

R n g Dom Rng Inv locatedIn Rng 3 116 1 1 6 116 116

SLIDE 18

Datasets & Metrics

• Data from Never-ending

Language Learner (NELL) from iteration 165

• Consists of over 1M

extractions and a rich

• ntology
• Evaluation set from (Jiang,

ICDM12) with 4.5K labeled extractions

• Report AUC-PR and running

time, optimization terms of slowest partition

Inputs Candidate Labels 1.2M Candidate Relations 100K Types Unique Labels 235 Unique Relations 221 Ontology Dom 418 Rng 418 Inv 418 Sub 288 RSub 461 Mut 17.4K RMut 48.5K

SLIDE 19

Experiments: Partitioning Approaches

Comparisons (6 partitions): NELL Default promotion strategy, no KGI KGI No partitioning, full knowledge graph model baseline KGI, Randomly assign extractions to partition Ontology KGI, Edge min-cut of ontology graph O+Vertex KGI, Weight ontology vertices by frequency O+V+Edge KGI, Weight ontology edges by inv. frequency

AUPRC Running Time

• Opt. Terms

NELL 0.765

• KGI

0.794 97 10.9M baseline 0.780 31 3.0M Ontology 0.788 42 4.2M O+Vertex 0.791 31 3.7M O+V+Edge 0.790 31 3.7M

SLIDE 20

Experiments: Scalability (Ont+Vertex)

0.794 0.791 0.791 0.790 0.788 0.5 0.55 0.6 0.65 0.7 0.75 0.8 0.85 0.9 0.95 1 10 20 30 40 50 60 70 80 90 100 1 2 3 6 12 24 48

Area Under Precision-Recall Curve Running Time (minutes) Number of Partitions

Partitions vs. Performance

Running Time

SLIDE 21

Conclusion

• Knowledge Graph Identification: a powerful technique for

constructing consistent knowledge graphs…

• Scalability is still a concern for real-world applications
• Partitioning can address scalability concerns
• Ontology-aware partitioning partitions the ontology instead of

the knowledge graph

• Including data distribution information balances partitions
• Results on the NELL dataset show this strategy reduces

running time from 97 minutes to 12 minutes without significant AUC degradation Code Available on GitHub: https://github.com/linqs/KnowledgeGraphIdentification