Semantic Relations within Data Sources Mohsen Taheriyan Craig A. - - PowerPoint PPT Presentation

▶
semantic relations within
SMART_READER_LITE
LIVE PREVIEW

Semantic Relations within Data Sources Mohsen Taheriyan Craig A. - - PowerPoint PPT Presentation

Jan 22, 2023 198 likes •495 views

Leveraging Linked Data to Discover Semantic Relations within Data Sources Mohsen Taheriyan Craig A. Knoblock Pedro Szekely Jose Luis Ambite Map Structured Data to Ontologies Map the source to the classes & properties in an ontology

slide-1
SLIDE 1

Leveraging Linked Data to Discover Semantic Relations within Data Sources

Mohsen Taheriyan Craig A. Knoblock Pedro Szekely Jose Luis Ambite

slide-2
SLIDE 2

Map Structured Data to Ontologies

Map the source to the classes & properties in an ontology

1

title date name 1 The Island 2009 Walton Ford 2 Excavation at Night 1908 George Wesley Bellows 3 Rose Garden 1901 Maria Oakey Dewing

Source Domain Ontology

CIDOC-CRM

slide-3
SLIDE 3

Semantic Types

E35_Title E82_Actor_Appellation

rdfs:label rdfs:label

2

E52_Time-Span title date name 1 The Island 2009 Walton Ford 2 Excavation at Night 1908 George Wesley Bellows 3 Rose Garden 1901 Maria Oakey Dewing

P82_at_some_time_within

slide-4
SLIDE 4

Relationships

E35_Title E82_Actor_Appellation

rdfs:label rdfs:label

3

E52_Time-Span title date name 1 The Island 2009 Walton Ford 2 Excavation at Night 1908 George Wesley Bellows 3 Rose Garden 1901 Maria Oakey Dewing

P82_at_some_time_within

E22_Man-Made_Object E12_Production E21_Person

P102_has_title P108_was_produced_by P4_has_time-span P14_carried_out_by P131_is_identified_by

slide-5
SLIDE 5

Problem:

How to automatically infer semantic relations?

slide-6
SLIDE 6

Idea

Exploit the relationships within already published linked data

5

slide-7
SLIDE 7

Extract schema-level graph patterns from LD

Approach

6

  • Target source (S)
  • Domain Ontologies (O)
  • Semantic labels of S
  • Linked Data (in the same domain)

Construct a graph from LD patterns and the ontology Generate and rank semantic models

1 2 3

Input

A ranked set of semantic models for S

Output

slide-8
SLIDE 8

Approach

7

Extract schema-level graph patterns from LD

  • Target source (S)
  • Domain Ontologies (O)
  • Semantic labels of S
  • Linked Data (in the same domain)

Construct a graph from LD patterns and the ontology Generate and rank semantic models

1 2 3

Input

A ranked set of semantic models for S

Output

slide-9
SLIDE 9

Schema-Level LD Patterns

8

../person- institution/57551 E21_Person rdf:type Thomas Burgon skos:prefLabel ../person- institution/57551/birth P98i_was_born ../person- institution/57551/birth/ date P4_has_time-span E67_Birth rdf:type 1787 rdfs:label E52_Time-Span rdf:type

LD fragment from the British Museum

slide-10
SLIDE 10

Schema-Level LD Patterns

9

E67_Birth E21_Person P98i_was_born

../person- institution/57551 E21_Person rdf:type Thomas Burgon skos:prefLabel ../person- institution/57551/birth P98i_was_born ../person- institution/57551/birth/ date P4_has_time-span E67_Birth rdf:type 1787 rdfs:label

E52_Time-Span P4_has_time-span

E52_Time-Span rdf:type

LD fragment from the British Museum Pattern

slide-11
SLIDE 11

Pattern Templates

  • Many possible templates for patterns

– Example: patterns for classes C1, C2, C3

  • Consider only tree patterns
  • Limit the length of the patterns

10

slide-12
SLIDE 12

Extracting LD Patterns

  • Use SPARQL to extract patterns of length
  • ne

11

Person

  • rganizer

Event location Place Event Place Person born

length 1

slide-13
SLIDE 13
  • Iteratively construct larger patterns by

joining with patterns of length 1

Extracting LD Patterns

12

Person

  • rganizer

Event location Place Place Event location born Person Person born Event

  • rganizer

Place Person born Place born Place Event location Place location Place Person

  • rganizer

Event

  • rganizer

Event

length 2

slide-14
SLIDE 14
  • Filter out the patterns not appearing in

the data

Extracting LD Patterns

13

Person

  • rganizer

Event location Place Place Event location born Person Person born Event

  • rganizer

Place Person born Place born Place Event location Place location Place Person

  • rganizer

Event

  • rganizer

Event

slide-15
SLIDE 15

Approach

14

Extract schema-level graph patterns from LD

  • Target source (S)
  • Domain Ontologies (O)
  • Semantic labels of S
  • Linked Data (in the same domain)

Construct a graph from LD patterns and the ontology Generate and rank semantic models

1 2 3

Input

A ranked set of semantic models for S

Output

slide-16
SLIDE 16

Merge the Patterns into a Graph

15 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

Start from longer patterns, skip the ones already in the graph

slide-17
SLIDE 17

Less weight for more popular links

W = (1 - freq)/(total count of links)

Weighting the Links

16 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

0.84 0.68 0.80 0.92 0.87 0.95 0.70 0.92

slide-18
SLIDE 18

Links from the same pattern have the same tag

Coherence

17 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

0.84 0.68 0.80 0.92 0.87 0.95 0.70 0.92 m1 m2 m2 m3 m3 m5 m5 m4

slide-19
SLIDE 19

High weights for links that do not have any instance in the data

Add the paths from the Ontology

18 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

0.84 0.68 0.80 0.92 0.87 0.95 0.70 0.92 m1 m2 m2 m3 m3 m5 m5 m4

P14_carried_out_by

100

slide-20
SLIDE 20

Approach

19

Extract schema-level graph patterns from LD

  • Target source (S)
  • Domain Ontologies (O)
  • Semantic labels of S
  • Linked Data (in the same domain)

Construct a graph from LD patterns and the ontology Generate and rank semantic models

1 2 3

Input

A ranked set of semantic models for S

Output

slide-21
SLIDE 21

Map Semantic Labels to the Graph

20 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

0.84 0.68 0.80 0.92 0.87 0.95 0.70 0.92 m1 m2 m2 m3 m3 m5 m5 m4

P14_carried_out_by

100

slide-22
SLIDE 22

Map Semantic Labels to the Graph

21 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

0.84 0.68 0.80 0.92 0.87 0.95 0.70 0.92 m1 m2 m2 m3 m3 m5 m5 m4

P14_carried_out_by

100

slide-23
SLIDE 23

Generate Semantic Models

22 E12_Production E53_Title P108i_was_produced_by E52_Time-Span E82_Actor_Appellation E22_Man-Made_Object E21_Person P102_has_title P131_is_identified_by E67_Birth P98i_was_born P4_has_time-span P4_has_time-span E39_Actor P14_carried_out_by P131_is_identified_by

0.84 0.68 0.80 0.92 0.87 0.95 0.70 0.92 m1 m2 m2 m3 m3 m5 m5 m4

P14_carried_out_by

100

  • Compute top k minimal trees
  • Consider both coherence and popularity
slide-24
SLIDE 24

Evaluation

23

Dataset Ontology Gold Standard Models Linked Data 29 museum data sources 458 attributes (columns) CRM 147 classes 409 properties 852 nodes 825 links RDF generated from the same dataset (leave-one-out) 29 museum data sources 458 attributes CRM 147 classes 409 properties 852 nodes 825 links RDF published by Smithsonian American Art Museum (more than 3 million triples) 29 museum data sources 329 attributes EDM 147 classes 409 properties 470 nodes 441 links RDF generated from the same dataset (leave-one-out) 15 sources containing data about weapon ads 175 attributes schema.org (ext) 736 classes 1081 properties 261 nodes 246 links RDF generated from the same dataset (leave-one-out)

slide-25
SLIDE 25

Example Gold Standard Models

24

slide-26
SLIDE 26
  • Compute precision and recall (between learned links

and correct links)

  • Correct semantic labels are given

Evaluation

25

Person Artwork location Museum creator

correct model

Person Museum location Artwork founder

learned model

<Artwork,location,Museum> <Artwork,creator,Person> <Museum,founder,Person> <Artwork,location,Museum>

Precision: 0.5 Recall: 0.5

slide-27
SLIDE 27

Results

26

max len of patter ns

Museum CRM (leave-one-

  • ut)

Museum CRM (Smithsonian LD) Museum EDM Weapon schema.org

precision recall precision recall precision recall precision recall

0.07 0.05 0.07 0.05 0.01 0.01 0.03 0.02 1 0.60 0.60 0.28 0.29 0.85 0.78 0.84 0.79 2 0.64 0.67 0.53 0.58 0.81 0.81 0.83 0.79

... ... ... ... ... ... ... ... ...

5 0.75 0.77 0.61 0.67 0.83 0.82 0.86 0.82

  • Very low accuracy if only using the ontology paths
  • Considering coherence improves the quality of the

models (longer patterns increase the accuracy)

  • Higher precision & recall for less complex ontologies
slide-28
SLIDE 28

Related Work

  • Understand semantics of Web tables [Wang et

al., 2012] [Limaye et al., 2010] [Venetis et al., 2011]

  • Link table values to the LOD entities [Muoz et

al., 2013] [Mulwad et al., 2013]

  • Learn semantic models from previously

modeled sources (Karma) [Taheriyan et al, 2015]

  • Extract schema-level patterns (SLPs, length
  • ne) from LOD [Schaible et al., 2016]

–E.g., ({Person,Player},{knows},{Person,Coach})

27

slide-29
SLIDE 29

Discussion

  • Manually constructing semantic models is

hard & expensive

–Needs domain knowledge and expertise in SW technologies –Often requires many user interactions in modeling tools

  • Infer semantic relations from linked data

–The suggested model can be refined in tools such as Karma

  • Help to publish consistent RDF data

28