t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt>...

48
WSDM 2018 J AY P UJARA AND S AMEER S INGH

Transcript of t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt>...

Page 1: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

WSDM 2018

JAY PUJARA AND SAMEER SINGH

Page 2: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Introducing Presenters

2

Jay Pujara: Research Scientist at USC/ISI

Sameer Singh: Assistant Professor at UCI

Page 3: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

3

https://kgtutorial.github.io

Page 4: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

4

Part 1: Knowledge Graphshttps://kgtutorial.github.io

Page 5: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

5

Part 2: Knowledge Extraction

Part 1: Knowledge Graphshttps://kgtutorial.github.io

Page 6: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

6

Part 2: Knowledge Extraction

Part 3:Graph Construction

Part 1: Knowledge Graphshttps://kgtutorial.github.io

Page 7: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

7

Part 2: Knowledge Extraction

Part 3:Graph Construction

Part 1: Knowledge Graphs

Part 4: Critical Analysis

https://kgtutorial.github.io

Page 8: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

8

Part 2: Knowledge Extraction

Part 3:Graph Construction

Part 1: Knowledge Graphs

Part 4: Critical Analysis

https://kgtutorial.github.io

Page 9: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

9

Part 2: Knowledge Extraction

Part 3:Graph Construction

Part 1: Knowledge Graphs

Part 4: Critical Analysis

https://kgtutorial.github.io

Page 10: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Outline1. Knowledge Graph Primer [Jay]

2. Knowledge Extraction Primer [Jay]

3. Knowledge Graph Constructiona. Probabilistic Models [Jay]

Coffee Break

b. Embedding Techniques [Sameer]

4. Critical Overview and Conclusion [Sameer]

10

Page 11: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What if I have a question?

11

Page 12: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Overview

12

Part 2: Knowledge Extraction

Part 3:Graph Construction

Part 1: Knowledge Graphs

Part 4: Critical Analysis

Page 13: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph PrimerTOPICS:

WHAT IS A KNOWLEDGE GRAPH?

WHY ARE KNOWLEDGE GRAPHS IMPORTANT?

WHERE DO KNOWLEDGE GRAPHS COME FROM?

KNOWLEDGE REPRESENTATION CHOICES

PROBLEM OVERVIEW

13

Page 14: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph PrimerTOPICS:

WHAT IS A KNOWLEDGE GRAPH?WHY ARE KNOWLEDGE GRAPHS IMPORTANT?

WHERE DO KNOWLEDGE GRAPHS COME FROM?

KNOWLEDGE REPRESENTATION CHOICES

PROBLEM OVERVIEW

14

Page 15: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?

15

Page 16: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?• Knowledge in graph form!

16

Page 17: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?• Knowledge in graph form!

• Captures entities, attributes, and relationships

17

Page 18: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?• Knowledge in graph form!

• Captures entities, attributes, and relationships

• Nodes are entities

18

E1

E2

E3

Page 19: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?• Knowledge in graph form!

• Captures entities, attributes, and relationships

• Nodes are entities

• Nodes are labeled with attributes (e.g., types)

19

E1

A1A2

E2

E3

A1A2

A1A3

Page 20: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?• Knowledge in graph form!

• Captures entities, attributes, and relationships

• Nodes are entities

• Nodes are labeled with attributes (e.g., types)

• Typed edges between two nodes capture a relationship between entities

20

E1

A1A2

E2

E3

A1A2

A1A2

Page 21: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Example knowledge graph• Knowledge in graph form!

• Captures entities, attributes, and relationships

• Nodes are entities

• Nodes are labeled with attributes (e.g., types)

• Typed edges between two nodes capture a relationship between entities

21

personplaceLiverpool

bandBeatles

John Lennon

Page 22: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph PrimerTOPICS:

WHAT IS A KNOWLEDGE GRAPH?

WHY ARE KNOWLEDGE GRAPHS IMPORTANT?WHERE DO KNOWLEDGE GRAPHS COME FROM?

KNOWLEDGE REPRESENTATION CHOICES

PROBLEM OVERVIEW

22

Page 23: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Why knowledge graphs?• Humans:

•Combat information overload•Explore via intuitive structure•Tool for supporting knowledge-driven tasks

• AIs:•Key ingredient for many AI tasks•Bridge from data to human semantics•Use decades of work on graph analysis

23

Page 24: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Applications 1: QA/Agents

24

Page 25: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Applications 2: Decision Support

25

Page 26: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Applications 3: Fueling Discovery

26

Page 27: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graphs & Industry•Google Knowledge Graph

• Google Knowledge Vault

•Amazon Product Graph•Facebook Graph API•IBM Watson•Microsoft Satori

• Project Hanover/Literome

•LinkedIn Knowledge Graph•Yandex Object Answer•Diffbot, GraphIQ, Maana, ParseHub, Reactor Labs,

SpazioDati27

Page 28: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph PrimerTOPICS:

WHAT IS A KNOWLEDGE GRAPH?

WHY ARE KNOWLEDGE GRAPHS IMPORTANT?

WHERE DO KNOWLEDGE GRAPHS COME FROM?KNOWLEDGE REPRESENTATION CHOICES

PROBLEM OVERVIEW

28

Page 29: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Where do knowledge graphs come from?

29

Page 30: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Where do knowledge graphs come from?• Structured Text

◦ Wikipedia Infoboxes, tables, databases, social nets

30

Page 31: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Where do knowledge graphs come from?• Structured Text

◦ Wikipedia Infoboxes, tables, databases, social nets

• Unstructured Text◦ WWW, news, social media,

reference articles

31

Page 32: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Where do knowledge graphs come from?• Structured Text

◦ Wikipedia Infoboxes, tables, databases, social nets

• Unstructured Text◦ WWW, news, social media,

reference articles

• Images

32

Page 33: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Where do knowledge graphs come from?• Structured Text

◦ Wikipedia Infoboxes, tables, databases, social nets

• Unstructured Text◦ WWW, news, social media,

reference articles

• Images

• Video◦ YouTube, video feeds

33

Page 34: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph PrimerTOPICS:

WHAT IS A KNOWLEDGE GRAPH?

WHY ARE KNOWLEDGE GRAPHS IMPORTANT?

WHERE DO KNOWLEDGE GRAPHS COME FROM?

KNOWLEDGE REPRESENTATION CHOICES

PROBLEM OVERVIEW

34

Page 35: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Simon&NewellGeneral Problem

Solver

McCarthyFormalizing

Commonsense

Hayes&McCarthyFrame Problem

QuillianSemantic Networks

ConceptNet

BrooksSubsumption

Minsky, FilmoreFrames

McCulloch&Pitts

Artificial Neurons

Minsky&Pappert

“Perceptrons”SystematicityDebate

2000 1990 1980 1970 1960 1950 19402000 1990 1980 1970 1960 1950 1940

SHRUTI

BobrowSTUDENT

WinogradSHRDLU

Rumelhart et alBackPropagationSeries of Neural-

Symbolic Models

Description Logic

LenantCyc

SLIDE COURTESY OF DANIEL KHASHABI

Page 36: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Representation•Decades of research into knowledge representation

•Most knowledge graph implementations use RDF triples• <rdf:subject, rdf:predicate, rdf:object> : r(s,p,o)• Temporal scoping, reification, and skolemization...

•ABox (assertions) versus TBox (terminology)

•Common ontological primitives• rdfs:domain, rdfs:range, rdf:type, rdfs:subClassOf, rdfs:subPropertyOf, ...• owl:inverseOf, owl:TransitiveProperty, owl:FunctionalProperty, ...

36

Page 37: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Semantic Web•Standards for defining and exchanging knowledge

• RDF, RDFa, JSON-LD, schema.org• RDFS, OWL, SKOS, FOAF

•Annotated data provide critical resource for automation

•Major weakness: annotate everything?

37"LINKING OPEN DATA CLOUD DIAGRAM 2014, BY MAX SCHMACHTENBERG, CHRISTIAN BIZER, ANJA JENTZSCHAND RICHARD CYGANIAK. HTTP://LOD-CLOUD.NET/"

Page 38: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Information Extraction from Text•Focus of this tutorial!

•Answer to the knowledge acquisition bottleneck

•Many challenges:• chunking• polysemy/word sense disambiguation• entity coreference• relational extraction

38

Page 39: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph PrimerTOPICS:

WHAT IS A KNOWLEDGE GRAPH?

WHY ARE KNOWLEDGE GRAPHS IMPORTANT?

WHERE DO KNOWLEDGE GRAPHS COME FROM?

KNOWLEDGE REPRESENTATION CHOICES

PROBLEM OVERVIEW

39

Page 40: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

What is a knowledge graph?• Knowledge in graph form!

• Captures entities, attributes, and relationships

• Nodes are entities

• Nodes are labeled with attributes (e.g., types)

• Typed edges between two nodes capture a relationship between entities

40

E1

A1A2

E2

E3

A1A2

A1A2

Page 41: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Basic problems

41

E1

A1A2

E2

E3

A1A2

A1A2

Page 42: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Basic problems

• Who are the entities (nodes) in the graph?

42

E1

A1A2

E2

E3

A1A2

A1A2

Page 43: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Basic problems

• Who are the entities (nodes) in the graph?

• What are their attributes and types (labels)?

43

E1

A1A2

E2

E3

A1A2

A1A2

Page 44: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Basic problems

• Who are the entities (nodes) in the graph?

• What are their attributes and types (labels)?

• How are they related (edges)?

44

E1

A1A2

E2

E3

A1A2

A1A2

Page 45: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Basic problems

• Who are the entities (nodes) in the graph?

• What are their attributes and types (labels)?

• How are they related (edges)?

45

E1

A1A2

E2

E3

A1A2

A1A2

Page 46: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Knowledge Graph Construction

46

Knowledge Extraction

Graph Construction

Page 47: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Two perspectivesKnowledge Extraction• Who are the entities

(nodes) in the graph?• Named Entity Recognition • Entity Coreference

• What are their attributes and types (labels)?• Named Entity Recognition

• How are they related (edges)?• Relation Extraction• Semantic Role Labeling

Graph Construction• Who are the entities

(nodes) in the graph?• Entity Linking• Entity Resolution

• What are their attributes and types (labels)?• Collective Classification

• How are they related (edges)?• Link Prediction

47

Page 48: t^ D î ì í ô W zh: Z E D Z /E', · 2018-02-05 · < v } Á o P ' Z W ] u dKW/ ^ W t, d/^ < EKt> ' ' Z W, M t,z Z < EKt> ' ' Z W,^ / DWKZd Ed M t, Z K < EKt>

Tutorial Outline1. Knowledge Graph Primer [Jay]

2. Knowledge Extraction Primer [Jay]

3. Knowledge Graph Constructiona. Probabilistic Models [Jay]

Coffee Break

b. Embedding Techniques [Sameer]

4. Critical Overview and Conclusion [Sameer]

48