Link to this document: https://short.upm.

es/odllz

Date: 31st of August (90 participants at 11:35)

Notes:

● Validating ontologies through requirements with Themis. (Alba

Fernández-Izquierdo - BASF)
○ Ontology verification compares the ontology against the
requirements.
○ http://themis.linkeddata.es/ → to test an ontology to check
whether it satisfies its functional requirements. Requirements are
ambiguous and difficult to formalize
○ Themis automates the implementation and execution of
requirements tests and also includes some inference additionally
to the defined test expressions
○ proposes both process and online tool suite (code shared under
Apache L2.0)/Swagger services for verification
○ an ontology for test requirements is provided
○ Possible Themis results: correct, missing terms, missing relations,
conflicting information
○ Several ontologies can be tested in parallel
○ GitHub repository at https://github.com/oeg-upm/themis
○ Offers only a part of the ontology evaluation - Themis check the
requirements but does not do model checks
○ Discussion on possible convergence with SHACL
○

● Ontology drafting in PURO (Vojtěch Svátek / Marek Dudáš - Prague

University of Economics and Business)
○ supports early development stages of ontologies and aims to
overcome “the blank canvas of OE”
○ Current approaches: (1) ad hoc diagrams and free text - difficult
to translate into ontological format; (2) graphics and controlled
NL that conform to an existing ontology language (e.g., OWL).
 ○ example of how a “simple” statement can be modeled in many
different ways/alternative encodings
○ Desiderata for drafting:
■ do not commit to a certain encoding, allow alternative
encodings
■ keep free of constraints of the encoding language
■ example-based
○ "PURO" is the essence of the ontological distinctions considered:
Particulars vs. Universals, and Relationships vs. Objects ; allowing
to keep the modeled reality “free of impurity” incurred by
language constraints
○ PURO components: (1) modeling languages - lightweight FOL
formalisation; (2) modeling guidelines - currently rudimentary; (3)
tooling for PURO model visual authoring, management
(modularisation/merging), transformation to OWL&OntoUML
○ Experiments: (1) PURO first, Protege next Vs. (2) Protege from start
for modeling the same textually described situation:
■ similar performance and correctness (but different errors)
■ lower user satisfaction for PURO
○ While developments ongoing, tool can be already used for
exploring ontology modeling challnegs
○ Puro Modeler: https://protegeserver.cz/purom5/#
○
● Reasonable Ontology Templates (OTTR) (Martin Georg Skjæveland -
University of Oslo)
○ Automates the task of having to read and interpret many PDF
documents to understand the terms
○ Presented AIBEL use case of a large industrial ontology
○ Requirements for building large ontologies (support large Nr of
classes, large nr of industry standards, consistent modeling,
collaborative development and automated mechanisms for
verification); Difficulties with existing ontology editors -e.g.
protege not suited for beginners in ontology engineering
 ○ Introduction of abstractions to improve the efficiency and quality
of ontology engineering, to capture reoccurring modeling
patterns
○ OTTR is a macro/pattern language and tool (avoids repetition,
separates design and content, encapsulates complexity, ensures
completeness)
■ The end user is only required to instantiate the template
and not deal with the high complexity expressions hidden
behind the template
■ Allows for easy updates of the used modeling patterns - all
instances get updated automatically once the template
gets and update
■ it can take tabular data as input
■ top-down modeling with different abstraction levels
○ It is impossible not to care about logic when building OWL
ontologies.
○ Templates library: http://tpl.ottr.xyz
■ Library management procedures defined and
maintenance tool available
○
● Democratisation of Enterprise Knowledge Graphs (EKG) (Katariina Kari -
IKEA)
○ Enterprise data challenges: no common view on data, data
storing, insights require complex sql queries, the data is not
connected to meaning
○ Solution to these challenges -EKG: 3 layers - (1) ontology - central
business concepts, (2) specialized instance layer - controlled
voc., taxonomies, defined in decentralized way and manually
created, (3) data graph, automatically created instances
■ adding a human-understanding layer on top of the data
○ interesting use of both ontologies and taxonomies in the same
setting
○ stakeholders of EKG: (1) Subject-matter experts that define the
data items interesting for their use case; (2) management - asks
high level questions; (3) applications; (4) ETL pipeline
 ○ industry concerns about EKG from industry:
■ reasoning (not needed in any known use case)
■ triple stores (not needed in any known use case)
■ ontology/taxonomy visualisation
● need of one solution for visualising both ontologies
and taxonomie
■ SPARQL
● need of query service with examples and
auto-complete (similar to Wikidata)
■ SHACL (not needed in any known use case)
■ top-ontology (not needed in any known use case)
○ Q&A:
■ GDPR leads to less people sharing personal data, so
collaborative filtering engines have reduced performance
due to data sparseness; this could be compensated with
explicitly declared knowledge in the ontology layer
● Methodology and tools for ontology development in BASF (Alba
Fernández-Izquierdo / Iker Esnaola - BASF)
○ Challenges: need for common terminology, dealing with
different meanings of the same term, redundant terms -> A
solution is to use ontologies
○ OMF (Ontology Management Framework) - goal is to reduce
effort and speed up reuse
○ Automate some tasks in the ontology development process
○ GOMO (Governance Operational Model for Ontologies) -
manage components of the ontology development process
○ OMF functionality: (1) support lifecycle for the ontology and the
ontology assets such as requirements, (2) ontology lookup, (3)
quality assurance, (4) publication of ontologies, their
documentation and their assets; the documentation can also be
generated automatically, (5) access control, (6) curation of
ontologies - identify errors or further requirements
■ tools and services provided (OMF pipeline)
■ automatic ontology workflow
 ○ Curation of existing ontologies: require collaboration between
ontology engineers and domain experts
■ roles: validators request updates, integrators implement
the changes
■ OntoMatik - automatic update service for ontologies, e.g.
collection of information from external sources, domain
experts‘ role is the validation rather than the
representation of the information
○
● OLS - The Ontology Lookup Service (Henriette Harmse - European
Bioinformatics Institute | EMBL-EBI)
○ ontology lookup service mostly for biomedical domain
○ 2 parts - a web app where you can browse ontologies(find terms,
individuals, download), text-based searches and traverse tees of
ontologies and an indexer for text and graph relations; 3rd part-
configurator
○ a rest api is also available for e.g. text-search in specific fields
○ https://github.com/ebispot/ols
○
● Building an Enterprise Ontology Management System for the
Pharmaceutical Industry (Simon Jupp - SciBite)
○ CENtree - ontology management platform; captures
provenance and provides a governance model
○ key challenge with versioning: how to make local changes to an
ontology but still be able to pull new changes from the public
ontology
○ different roles in the governance model: user, suggester,editor
and admin; the same user can have different roles in different
ontologies; only admin can accept or reject changes
○ available conflict detection and resolution +manager when
merging conflicting changes
● Declarative Knowledge Graph Construction: A practical introduction
(David Chaves-Fraga - Universidad Politécnica de Madrid - KULeuven)
○ KG construction is often done by scripting - but this is not a
sustainable approach (issues: efficiency, scalability …)
 ○ Alternative is the KG construction with declarative mappings
between ontologies and data sources
■ nice historic overview of declarative mapping languages
for schema transformations, from XSPARQL to ShExML
○ Morph-KG offers scalable KG construction based on (R2)RML
○ Tutorial guide:
https://colab.research.google.com/drive/1ByFx_NOEfTZeaJ1Wtw
3UwTH3H3-Sye2O?usp=sharing
○ http://davidchavesfraga.com/outcomes/presentations/2022/de
clarative2022ontocommons.pdf
● Applying Ontotext GraphDB and Ontotext Refine for building industrial
Knowledge Graphs (Miroslav Chervenski / Vladimir Alexiev - Ontotext)
○ Overview of OntoText GraphDB
○ Various GraphDB access and interface methods available
○ Also facilitates Ontotext-developed plugins or your
own/third-party ones via their architecture
○ Many significant users in enterprise/industry, government, etc.
○ OntoText Refine based on OpenRefine, command line and API
support
● MatVis - A Framework to Visually Represent Material Science
Engineering Methods and RDF Knowledge Graph Creation (Andre
Valdestilhas - BAM Bundesanstalt für Materialforschung)
○ There is a strong need for visual graph creation in the material
science domain, as a lot of manual work looking at publications,
seeing if data is suitable or conforms to any standards, can be
extracted, etc.
○ Data is obtained, a visual version is created, a KG is then made
from this
○ Being used in a variety of projects between BAM and Fraunhofer
○ Best way to visual represent material science (engineering)
methods is with a directed graph
○ Reproducibility/reusability of experiments and obtained data is
the major problem from users
 ○ Interactions between the current approach and end users who
are not ontologists but material science domain experts remain
to be discussed
● Health application (Jaan Altosaar - One Fact Foundation)
○ The definition of Knowledge Graph should be refined. Because
when thinking about how the KGs are created and by who
(human or machine/robot), the quality of the KG can be
questionable
○ NLP and human annotation are used in generating KG in
medicine, but trust issue of human annotation remain as an issue
○ Auditing KG quality methods: value chain analysis, ethnography,
statistical analysis, ML and data science, open source
intelligence reporting
○ Level of trust of data in health care from low level to high level:
patients, providers, health information exchanges, payors,
pharma companies, regulations from governments; Value chain
in KGs in health care has the inverse order
○ Data annotation in medical domain costs a fortune
○ Uptodate: https://archive.org/details/Uptodate21.6
○ Slides:
https://www.dropbox.com/s/ih8r0u2rid77thn/220831-onefact.org
-ontocommons.eu-quality-assessment-knowledge-graphs.pdf?dl
=0
○
● Open Semantic Lab - Bringing ontologies into everyday science (Simon
Stier - Fraunhofer)
○ https://github.com/OpenSemanticLab
○ https://kiprobatt.de/wiki/Main_Page
○
General comments, lessons learnt, takeaways:
● Priorities:
○ Is industry highly interested in or prioritizing ontology
search? information retrieval
○ change management?
○ visualization?