Is knowledge engineering
still relevant?
Mathieu d’Aquin - @mdaquin
mathieu.daquin@insight-centre.org
Summary of the talk: yeah, probably...
A quick something about me
2002-2006
2006-2017
2017-now
Knowledge representation,
case-based reasoning,
oncology
Ontology engineering,
semantic web, linked data,
smart cities, education,
privacy
Data science, data analytics,
data and knowledge
engineering, data ethics,
knowledge graphs
knowledge engineering
?
Knowledge-based systems
Knowledge-based
system
Input
Question
Problem
Situation
Answer
Inference
Solution
Decision
Knowledge
Base
Knowledge engineering
Knowledge-based
system
Input
Question
Problem
Situation
Answer
Inference
Solution
Decision
Knowledge
Base
Design, representation,
implementation,
manipulation, curation
and processing of this
Example (oncology)
M. d’Aquin et al., Knowledge editing and
maintenance tools for a semantic portal in
oncology, International journal of
human-computer studies 62 (5), 2005
But that’s old stuff, right?
Knowledge
base
Reasoning
Oncology
expert
Knowledge
engineer
Patient
data
Treatment
Manual effort
Classification
Machine
learning model
Training
Lots of data about
patients and treatments
Patient
data
Treatment
So what else did we use knowledge engineering for?
Making knowledge available and machine
processable on the Web, i.e. the Semantic Web
Ontologies: Knowledge representation with
concepts and relations.
Web ontologies: Ontologies that live well on
the web.
Linked data: Data as a graph of web entities,
and where the elements (nodes and edges) are
defined by web ontologies.
Knowledge graphs: See linked data ;)
Shameless plug
The semantic web
Network of
knowledge artefact
Linked Data
Network of data
artefact
The Web
Network of
documents
The Internet
Network of machines
Example
Annotating correspondances from early women philosophers
(PhD of Ioana Kyvernitou)
Example
Annotating correspondances from early women philosophers
(PhD of Ioana Kyvernitou)
Knowledge engineering for the meta-aspects of Data Science
Hypo. /
Question
Plan Collect
data
Analyse
data
Extract
results
Exploit
results
Data Models
New
info
What-
ever
was the
goal
Anything new?
Knowledge engineering for the meta-aspects of Data Science
Hypo. /
Question
Plan Collect
data
Analyse
data
Extract
results
Exploit
results
Data Models
New
info
What-
ever
was the
goalDataset
License
Regulation
Source
Dataset
Characteristics
Data
Science
Task
Technique
Model
Model
Parameters
...
associated with
obtained from with
derived from
used for
implemented by
using
produced
version of
produced
Propagating data policies
Propagation of policies in rich data flows
E Daga, M d'Aquin, A Gangemi, E Motta,
International Conference on Knowledge
Capture, K-CAP 2015
Explaining data patterns
Dedalo: Looking for clusters explanations in
a labyrinth of linked data, I Tiddi, M
d’Aquin, E Motta European Semantic Web
Conference, 2014
An ontology design pattern to define
explanations, I Tiddi, M d'Aquin, E Motta
Proceedings of the 8th International
Conference on Knowledge Capture, 2015
Understanding the link between data and what is done with it
(PhD Mohamed Adel)
Understanding the link between data and what is done with it
(PhD Mohamed Adel)
Making technological artefacts more non-expert friendly
An ontology-based approach to improve the
accessibility of ROS-based robotic systems
I Tiddi, E Bastianelli, G Bardaro, M d'Aquin, E
Motta, Knowledge Capture Conference, 2017
Extracting robot’s capabilities through automatically annotating components of
ROS (the Robot Operating System) with an ontology
Making technological artefacts more non-expert friendly
An ontology-based approach to improve the
accessibility of ROS-based robotic systems
I Tiddi, E Bastianelli, G Bardaro, M d'Aquin, E
Motta, Knowledge Capture Conference, 2017
What about making research data more
non-data expert friendly?
Ontologies to represent the base capabilities of datasets
Mapping data access to basic ontology operations
[{ "datasetId": "tempo",
"load":
"pd.read_csv('/home/mdaquin/data/TempoData/Beethoven_Op57_Tempi.csv')",
"types": {
"performance": {
"list": "[{'music': 'Bethoveen Op57', 'name': s.lower()} for s in
list(data.columns.values)[:-2]]",
"attributes": "{'music': 'music', 'name': 'value', bars:
'[peformedBar]'}",
"values": "if attribute=='bars':n result=[]n col=0n for
x,coln in enumerate(list(data.columns.values)[:-2]):n if
coln.lower()==obj['name']:n col=colnn for i in range(0,
len(data[col])):n result.append({'bar': {'music': 'Bethoveen Op57',
'number': i}, 'performance': {'music': 'Bethoveen Op57', 'name':
obj['name']}})"
},
"performedBar": {
"list": "[{'bar': {'music': 'Bethoveen Op57', 'number':m},
'performance': {'music': 'Bethoveen Op57', 'name': n.lower()}} for m in
list(range(0,data.shape[0])) for n in list(data.columns.values)[:-2]]",
"attributes": "{'bar': 'bar','performance': 'performance','tempo':
'value'}",
"values": "result=['unknown']nif attribute=='bar':n result =
[obj['bar']]nif attribute=='performance':n
result=[obj['performance']]nif attribute=='tempo':n col=0n for x,coln
Ontology-based data access in scratch
Ontology-based data access in scratch
Ontology-based data access in scratch
Ontology-based data access in scratch
Ontology-based data access in scratch
Conclusion
Knowledge is, more than ever, a necessary, valuable asset, and
some amounts of knowledge engineering can (and is) useful in
event the most top-down, data-centric of processes.
Need to not only scale, but also make more accessible, and more
integrated the tools to enable knowledge curation,
knowledge-based explanations/interpretation, and
knowledge-driven data access, integration and interpretation.
One size does not fit all: querying web
polystores, Y Khan, A Zimmermann, A Jha, V
Gadepally, M D’Aquin, R Sahay
Ieee Access 7, 2019
Towards an ethics by design methodology
for AI research projects, M d'Aquin, P
Troullinou, NE O'Connor, A Cullen, G Faller,
L Holden AAAI/ACM Conference on AI,
Ethics, and Society, 2018
Crowdsourcing Linked Data on listening
experiences through reuse and
enhancement of library data, A Adamou et
al. International Journal on Digital
Libraries 20 (1), 2019

Is knowledge engineering still relevant?

  • 1.
    Is knowledge engineering stillrelevant? Mathieu d’Aquin - @mdaquin mathieu.daquin@insight-centre.org Summary of the talk: yeah, probably...
  • 2.
    A quick somethingabout me 2002-2006 2006-2017 2017-now Knowledge representation, case-based reasoning, oncology Ontology engineering, semantic web, linked data, smart cities, education, privacy Data science, data analytics, data and knowledge engineering, data ethics, knowledge graphs
  • 3.
  • 4.
  • 5.
  • 6.
    Example (oncology) M. d’Aquinet al., Knowledge editing and maintenance tools for a semantic portal in oncology, International journal of human-computer studies 62 (5), 2005
  • 7.
    But that’s oldstuff, right? Knowledge base Reasoning Oncology expert Knowledge engineer Patient data Treatment Manual effort Classification Machine learning model Training Lots of data about patients and treatments Patient data Treatment
  • 8.
    So what elsedid we use knowledge engineering for? Making knowledge available and machine processable on the Web, i.e. the Semantic Web Ontologies: Knowledge representation with concepts and relations. Web ontologies: Ontologies that live well on the web. Linked data: Data as a graph of web entities, and where the elements (nodes and edges) are defined by web ontologies. Knowledge graphs: See linked data ;)
  • 9.
    Shameless plug The semanticweb Network of knowledge artefact Linked Data Network of data artefact The Web Network of documents The Internet Network of machines
  • 10.
    Example Annotating correspondances fromearly women philosophers (PhD of Ioana Kyvernitou)
  • 11.
    Example Annotating correspondances fromearly women philosophers (PhD of Ioana Kyvernitou)
  • 12.
    Knowledge engineering forthe meta-aspects of Data Science Hypo. / Question Plan Collect data Analyse data Extract results Exploit results Data Models New info What- ever was the goal
  • 13.
    Anything new? Knowledge engineeringfor the meta-aspects of Data Science Hypo. / Question Plan Collect data Analyse data Extract results Exploit results Data Models New info What- ever was the goalDataset License Regulation Source Dataset Characteristics Data Science Task Technique Model Model Parameters ... associated with obtained from with derived from used for implemented by using produced version of produced
  • 14.
    Propagating data policies Propagationof policies in rich data flows E Daga, M d'Aquin, A Gangemi, E Motta, International Conference on Knowledge Capture, K-CAP 2015
  • 15.
    Explaining data patterns Dedalo:Looking for clusters explanations in a labyrinth of linked data, I Tiddi, M d’Aquin, E Motta European Semantic Web Conference, 2014 An ontology design pattern to define explanations, I Tiddi, M d'Aquin, E Motta Proceedings of the 8th International Conference on Knowledge Capture, 2015
  • 16.
    Understanding the linkbetween data and what is done with it (PhD Mohamed Adel)
  • 17.
    Understanding the linkbetween data and what is done with it (PhD Mohamed Adel)
  • 18.
    Making technological artefactsmore non-expert friendly An ontology-based approach to improve the accessibility of ROS-based robotic systems I Tiddi, E Bastianelli, G Bardaro, M d'Aquin, E Motta, Knowledge Capture Conference, 2017 Extracting robot’s capabilities through automatically annotating components of ROS (the Robot Operating System) with an ontology
  • 19.
    Making technological artefactsmore non-expert friendly An ontology-based approach to improve the accessibility of ROS-based robotic systems I Tiddi, E Bastianelli, G Bardaro, M d'Aquin, E Motta, Knowledge Capture Conference, 2017
  • 20.
    What about makingresearch data more non-data expert friendly?
  • 21.
    Ontologies to representthe base capabilities of datasets
  • 23.
    Mapping data accessto basic ontology operations [{ "datasetId": "tempo", "load": "pd.read_csv('/home/mdaquin/data/TempoData/Beethoven_Op57_Tempi.csv')", "types": { "performance": { "list": "[{'music': 'Bethoveen Op57', 'name': s.lower()} for s in list(data.columns.values)[:-2]]", "attributes": "{'music': 'music', 'name': 'value', bars: '[peformedBar]'}", "values": "if attribute=='bars':n result=[]n col=0n for x,coln in enumerate(list(data.columns.values)[:-2]):n if coln.lower()==obj['name']:n col=colnn for i in range(0, len(data[col])):n result.append({'bar': {'music': 'Bethoveen Op57', 'number': i}, 'performance': {'music': 'Bethoveen Op57', 'name': obj['name']}})" }, "performedBar": { "list": "[{'bar': {'music': 'Bethoveen Op57', 'number':m}, 'performance': {'music': 'Bethoveen Op57', 'name': n.lower()}} for m in list(range(0,data.shape[0])) for n in list(data.columns.values)[:-2]]", "attributes": "{'bar': 'bar','performance': 'performance','tempo': 'value'}", "values": "result=['unknown']nif attribute=='bar':n result = [obj['bar']]nif attribute=='performance':n result=[obj['performance']]nif attribute=='tempo':n col=0n for x,coln
  • 24.
  • 25.
  • 26.
  • 27.
  • 28.
  • 29.
    Conclusion Knowledge is, morethan ever, a necessary, valuable asset, and some amounts of knowledge engineering can (and is) useful in event the most top-down, data-centric of processes. Need to not only scale, but also make more accessible, and more integrated the tools to enable knowledge curation, knowledge-based explanations/interpretation, and knowledge-driven data access, integration and interpretation. One size does not fit all: querying web polystores, Y Khan, A Zimmermann, A Jha, V Gadepally, M D’Aquin, R Sahay Ieee Access 7, 2019 Towards an ethics by design methodology for AI research projects, M d'Aquin, P Troullinou, NE O'Connor, A Cullen, G Faller, L Holden AAAI/ACM Conference on AI, Ethics, and Society, 2018 Crowdsourcing Linked Data on listening experiences through reuse and enhancement of library data, A Adamou et al. International Journal on Digital Libraries 20 (1), 2019