ROC: a method for proto-ontology construction by domain experts

1
ROC a method for proto-ontology construction by
domain experts

• Nicole Koenderink
• Lars Hulzebos
• Mark van Assem

Jeen Broekstra Remko van Brakel Jan Top
2
Outline

• Problem context
• Proposed solution ROC
• What is it?
• Related work
• How does it work?
• How well does it work?
• Conclusions and future work

Nicole Koenderink, AF--IM
2
3
Problem context / Motivation

• How to automate a knowledge-intensive task?
• Automation of quality assessment of seedlings
• Support of food researchers workflow
• Emerging risk assessment
• Healthy food components disclosure
• ? Create and use an ontology ?

Nicole Koenderink, AF--IM
3
4
Problem context / Motivation

• Difficulties in creating ontologies
• Technical modelling constructs, formal languages
  and KM-tools are hard to understand for domain
  experts
• Elicitating concepts is time-consuming (for
  domain expert and knowledge engineer)
• Maintaining focus is difficult

Nicole Koenderink, AF--IM
4
5
Solution every man to his trade

• ROC supports the Domain Expert
• Strength of a domain expert
• gathering task-relevant knowledge,
• not behaving as a knowledge engineer!

Nicole Koenderink, AF--IM
5
6
Overview

• Problem context
• Proposed solution ROC
• What is it?
• Related work
• How does it work?
• How well does it work?
• Conclusions and future work

Nicole Koenderink, AF--IM
6
7
ROC Reuse-based Ontology Construction

• ROC is a method to guide the domain expert in
  defining a proto-ontology
• A proto-ontology is a structured list of relevant
  triples. This is the basis for the final
  ontology.
• ROC reuses existing knowledge gathered from
  thesauri, ontologies, glossaries, etc.

Nicole Koenderink, AF--IM
7
8
ROC related work

• Methontology
• Watson-plugin for NeOn

Nicole Koenderink, AF--IM
8
9
Methontology
Methontology (Fernandez, 1997)
10
ROC and Methontology

• ROC builds upon Methontology
• Specification is used by domain expert and
  problem owner for setting the scope
• Specification is used by ROC to identify reusable
  sources
• Integration is brought forward in the process
  reusable sources are identified and actively used

Nicole Koenderink, AF--IM
10
11
ROC and Watson plugin for NeOn

• Both
• Reuse related RDF-triples from the Internet
• Difference
• Watson plug-in is aimed at knowledge engineer
  use of KM-tool, full RDF-context.
• ROC is aimed at domain expert only terms are
  shown for approval/rejection.

Nicole Koenderink, AF--IM
11
12
Overview

• Problem context
• Proposed solution ROC
• What is it?
• Related work
• How does it work?
• How well does it work?
• Conclusions and future work

Nicole Koenderink, AF--IM
12
13
ROC -- setup
Specify purpose
Identifysources
Extracttriples
Sesame repository
Proto-ontology
Nicole Koenderink, AF--IM
13
14
ROC statement repository

• Content harvested from multiple sources
• Thesauri, glossary, ontologies Agrovoc, Nalt,
  Cabi,
• Semi-structured documents Mathworld, Wikipedia,
• Information available through SPARQL queries

Sesame repository
14
15
ROC -- setup
Specify purpose
Identifysources
Extracttriples
Sesame repository
ROC Tool
Proto-ontology
Nicole Koenderink, AF--IM
15
16
ROC method
Nicole Koenderink, AF--IM
16
17
ROC -- steps
Nicole Koenderink, AF--IM
17
18
Example food ontology
19
ROC -- steps
Nicole Koenderink, AF--IM
19
20
Example food ontology
21
Example food ontology
22
ROC -- steps
Nicole Koenderink, AF--IM
22
23
Example food ontology
Nicole Koenderink, AF--IM
23
24
ROC -- setup
Specify purpose
Identifysources
Extracttriples
Sesame repository
ROC Tool
Proto-ontology
RDF-file
Nicole Koenderink, AF--IM
24
25
Example food ontology
Nicole Koenderink, AF--IM
25
26
Overview

• Problem context
• Proposed solution ROC
• What is it?
• Related work
• How does it work?
• How well does it work?
• Conclusions and future work

Nicole Koenderink, AF--IM
26
27
ROC -- evaluation

• Initial evaluation with respect to objectives
• Is ROC efficient for the domain expert?
• Is the proto-ontology useful for the specified
  task?
• Does the domain expert benefit from the reuse of
  existing sources?
• Full evaluation awaits final phase of ROC

Nicole Koenderink, AF--IM
27
28
ROC evaluation Domain expert support

• Comparison with interview-based ontology
  construction
• supply chains ontology (ROC-based)
• potato ontology (interview-based)
• Conclusion
• Ratio domain expert/knowledge engineer is better
  for ROC
• Individual times and total time is better for ROC

29
ROC -- evaluation

• Initial evaluation with respect to objectives
• Is ROC efficient for the domain expert?
• Is the proto-ontology useful for the specified
  task?
• Does the domain expert benefit from the reuse of
  existing sources?

Nicole Koenderink, AF--IM
29
30
ROC evaluation Task dedication

• Domain supply chains
• Expert finder based on proto-ontology
• Appropriate experts were indeed found

31
ROC -- evaluation

• Initial evaluation with respect to objectives
• Is ROC efficient for the domain expert?
• Is the proto-ontology useful for the specified
  task?
• Does the domain expert benefit from the reuse of
  existing sources?

Nicole Koenderink, AF--IM
31
32
ROC evaluation Reuse of existing sources

• Supply chain ontology
• Sources CABI, Agrovoc, Wikipedia, MeSH, NALT,
  UMLS, two natural language reports
• Reuse see table
• Conclusion many different sources were used

33
Overview

• Problem context
• Proposed solution ROC
• What is it?
• Related work
• How does it work?
• How well does it work?
• Conclusions and future work

Nicole Koenderink, AF--IM
33
34
Conclusions

• ROC helps to easily create a domain model
• Domain expert is properly supported
• Existing sources are actively reused

Nicole Koenderink, AF--IM
34
35
Future work

• Role of knowledge engineer from triples to
  proper ontology
• Domain expert support multiple expert mode,
  quality dashboard
• Use of Machine Learning and Text Mining methods
  for better reuse of existing sources

Nicole Koenderink, AF--IM
35
36
Thank you for your attention
Questions? Remarks?
Nicole.Koenderink_at_wur.nl http//www.afsg.nl/IM