We provide our own thesaurus in SKOS (
https://zbw.eu/stw/version/latest/download/about.en.html) but within that SKOS file we have some more modeling choices that are rather specific: The file includes both descriptors and a classification with codes for the top level subthesauri, which you can see here:
https://zbw.eu/stw/version/latest/about.en.htmlNow when training a model we want the model to only choose/target descriptors but not subthesauri notations so right now we exclude those by hand by modifying one line in our local Annif code. We could also just eliminate those codes from the intermediate format and then use that, and there are other options.
Our question to the community: Are there other users among you going the SKOS route and modelling their knowledge organization systems that way facing similar issues so that it would be an added value for the whole community to make it configurable which specific elements within a SKOS vocabulary are considered admissible candidates and which are not?
(I know this sounds convoluted but my hope is that institutions curating their own SKOS vocabularies will know roughly what I am talking about.)