knowledge sharing approach: RDA's I-ADOPT (InteroperAble Descriptions of Observable Property Terminologies)

[StroemPhi]

This issue is associated with the charter epic #4.

What cross-domain knowledge sharing approach does this issue address?

See report of the homonymous RDA WG: https://doi.org/10.15497/RDA00071
I-ADOPT is "a framework to harmonise the way observable properties are named and conceptualised, in various communities within and across scientific domains. There was a realization that the rapid demand for controlled vocabularies specialised in describing observed properties (i.e. measured, simulated, counted quantities, or qualitative observations) was presenting a risk of proliferation of semantic resources that were poorly aligned. This, in turn, was becoming a source of confusion for the end-users and a hindrance to data interoperability.

The development of the I-ADOPT Framework proceeded in multiple phases. Following the initial phase dedicated to the collection of user stories primarily from the environmental domain, the identification of key requirements, and an in-depth analysis of existing semantic representations of scientific variables and of terminologies in use, the group focused on identifying the essential components of the conceptual framework, reusing as much as possible concepts that were common to existing operational resources. The proposed framework was then tested against a variety of examples to ensure that it could be used as a sound basis for the creation of new variable names as needed. The results were formalised into the I-ADOPT ontology and subsequently extended with usage guidelines to form the I-ADOPT Framework presented in this document. The output can now be used to facilitate interoperability between existing semantic resources and to support the provision of machine-actionable variable descriptions whose components are mapped to FAIR vocabulary concepts. The easier integration of datasets annotated with I-ADOPT-enabled variable descriptions will not only enable more comprehensive analyses, but also provides the corpora for machine learning and other artificial intelligence applications."

Outcomes of previous discussions around I-ADOPT within our WG

• first discussed based on a presentation by Naouel Karam in our 2023-04-12 call
  • lead to an abstract submission to CORDI 2023: https://doi.org/10.52825/cordi.v1i.366
  • There was mild agreement that using I-ADOPT-based knowledge graphs for datasets within the NFDI would help in the interdisciplinary findability of datasets that probed/observed the same or similar entities.
  • Open questions remain around the applicability of I-ADOP in the humanities, where the research variable (object of interest) is rather often immaterial

What further steps are needed to be taken or discussed by/in our WG regarding this issue?

1. Collect more examples from multiple domains/consortia that use I-ADOPT to describe the observed variable which lead to a specific dataset
2. Participate within the I-ADOPT Variable Modeling Challenge, from September 16 to 22, 2024.

[dalito]

The flexibility of I-ADOPT allows different decisions about how to model variables. For example, let´s say we have a chemical reactor with an exothermic reaction and want to express the temperature increase. So we have:

• the target variable "temperature difference": T_delta
• the inlet temperature T_wall
• the temperature inside the reactor: T_reaction

Even in this simple example different decisions may be made:

• The reactor, is it a matrix or context?
• What is the minimum context? e.g. instance (the specific reactor) or class (the reactor type) or both?
• How to express that T_delta is a derived variable? Should T_wall, T_reaction be "context"?
• Should all three variables form a variable set?

These questions should have one answer else interoperability is limited. So to make I-ADOPT really useful, one also needs:

• A curated collection of variables expressed in I-ADOPT for the specific domain. This collection of variables indirectly provides the answers to the modelling questions.

or

• A detailed semantic model that guides the modelling of variables in the domain. From this more detailed model, a (simpler) I-ADOPT based model could be derived automatically and therefore unambiguously.