1. Construction of Enterprise Knowledge Graphs
Chapter 4

3. Outline Knowledge graph lifecycle Ontology authoring
Semi-atomated linking of Enterprise Data for Virtual knowledge graph
Fokus på å lage knowledge graphs med menneskelig innblanding

4. Outline Knowledge graph lifecycle Ontology authoring
Semi-atomated linking of Enterprise Data for Virtual knowledge graph
Fokus på å lage knowledge graphs med menneskelig innblanding

5. A general lifecycle

6. 1 Specification Draw up detailed specification One of the main tasks
1) Identification and analysis of data sources
2) URI design
Select data to integrate and publish
Data that exists in the organization
Needed external data

7. URI design Put as much information into the URI as possible
<
Use slash instead of hash URI whenever possible
Separate TBox (ontology model) from ABox (instances)
TBox: Append ”ontology” to base URI (ontology/Person)
ABox: Append ”resource” to base URI (resource/Erna)

8. (2) Modelling
Determine ontology to be used for modelling of the domain
Reuse as much as possible
If no suitable ontology is found, reuse parts
If nothing works out, start from scrach (follow NeOn methodology)

9. A general lifecycle

10. (3) Data lifting Transfer existing data to RDF Two main activities
Transformation
Linking
GRDDL, RDBS2RDF

11. Transformation Requirements
Full conversion – queries on original data source must be possible on RDF version
RDF instances should reflect target ontology structure (as closely as possible)
RDB2RDF, GRDDL, Google Refine/OpenRefine (RDF extension), D2R Server, ODEMapster, Stats2RDF

12. Linking Create links between our knowledge graph and external graphs
Steps:
1. Identify KG’s that are suitable as linking targets - manual
2. Discover relationships between items in our- and external KG – tools exists
3. Validate relationships – performed by domain experts
Finnes lager med KG’s på ”Linked Data repositories” som CKAN – manuelt

13. 4 Data publication Activities Knowledge graph publication
Metadata publication

14. Knowledge graph publication
Store and publish RDF data
Virtuoso Universal Server
Jena
Sesa
4Store
YARS
Some already include SPARQL endpoints

15. Metadata publication Include metadata information about the KG
Data about structure
Data about access
Descirption of links between knowledge graphs

16. A general lifecycle

17. Data Curation
Aims at maintaining and preserving data for reuse over time
Cleaning noise
Identify errors
(40x/50x errors)
Broken links
Malformed data types (”true” as xsd:int)
Bevaring

18. Outline Knowledge graph lifecycle Ontology authoring
Semi-atomated linking of Enterprise Data for Virtual knowledge graph
Fokus på å lage knowledge graphs med menneskelig innblanding.

19. Ontology Authoring - A compentency question-driven approach
Real-world ontologies requires manual constructions
Requires deep and complex professional knowledge
Onthology authors are domain experts not KG experts
Onthology authoring is time-consuming and error prone
Solution: ”Competency question-driven ontology authoring” (CQOA)

20. Competency Questions
Ontology must be able to answer competency questions (CQ)
Natural language sentences
Semiformal pattern: ”Which [CE1][OPE][CE2]?”
Examples:
”Which mammals eat grass?” (animal ontology)
Which processes implement an algorithm” (Software engineering ontology)
CQs are especially helpful to ontology authors

21. Presuppositions
”A special condition that must be met for a linguistic expression to have a denotation”
Example:
”Which processes implement an algorithm?”
Ontology must satisfy the following presuppositions:
Classes ”Process”, ”Algorithm” and property ”Implements” occurs in ontology
Ontology allows ”Process” to implement ”Algorithm”
Ontology allows ”Process” to not implement ”Algorithm”

22. Formulation of competency questions
Selection: ”Which mammals eat grass?”
Binary: Should answer the question with a boolean value (yes/no)
Counting question: Should answer with a number. ”How many pizzas has ham or chicken as topping?”
Question Polarity: ”Which pizza has no vegetables?”
Predicate arity: ”Is it thin or thick bread?”
Modifier: ”If I have 3 ingredients, how many pizzas can I make?”
Selection question
Binary question
Counting question
Question Polarity
Predicate Arity
Modifier

23. Test suite of CQs
Table 4.1 (p. 99)

24. Outline Knowledge graph lifecycle Ontology authoring
Semi-atomated linking of Enterprise Data for Virtual knowledge graph
Fokus på å lage knowledge graphs med menneskelig innblanding

25. Semi-automated linking of Enterprise Data for knowledge graphs
Activity is part of the ”Data lifting” step in the life cycle
Create data linkage
Helix: linking information sources
Build a knowledge graph for data discovery

26. Techniques of data discovery
Normalize data in different format
Index structured data in tables
Perform semantic matching between schema elements of structured data
Tag data with semantic tags
Find linkage points in the data so that users can join between tables

27. Helix input sources
Semi-structured sources (API / RDBMS, triple stores)
Online or local file stores
Online web API’s

28. Helix pre-processing Implemented in the HADOOP ecosystem
1. Schema discovery
2. Full-text indexing
3. Linkage discovery
Output: Semantically tagged Global Schema Graph

29. Linkage discovery
All-to-all instanced based matching of all attributes
Does not scale
Turn the problem into IR-problem

30. Linkage discovery example
Si noe om skoler som hadde stemmelokaler.
I NY brukte kan KG til å finne fram til sykehus ved hjelp av graf-traversering i stedet for fritekst-søk.