Repository navigation
How should NeoWiki data be projected into standard ontologies like CIDOC-CRM and EDM? (OntologyMapping.md) #996
Replies: 2 comments 3 replies
|
Here is a list of additional questions to the OntologyMapping.md doc that may need more discussion within the consortium as a whole: Q1: Aren’t multiple triple stores (for multiple ontologies) an extra burden for the users? Q2: 1-to-1 mapping (native schema to standard ontology) seems strange in the context of the semantic web; most KGs I know employ 1-to-many mappings (and utilise concepts like close or broadMatch); multiple mappings should be possible to query in the same triple store; esp. with help from agents. Q3: Local properties - still unclear how does multilinguality work (maybe this is described elsewhere)? and how does general referencing (of said properties) work (for users) - on internal (within the wiki instance) & external level (from wiki to other systems)? Q4: Biggest problem of users in Wikibase is not the data model (if anything it’s too flexible), but rather the lack of standards that can “just-work” out of the box. The process of data modelling is number 1 issue new WB users report (confirmed over 10 years worth of workshops and community meetings). How does NeoWiki address this? Q5: Could we opt for the possibility (at least) to also use native RDF - e.g. using a triple value IRI from another semantic web resource without duplicating it. Wikidata have some experiments going on now with federating entities (at least between Wikibase instances) Q6: Are there any arguments against event-centric modelling being implemented via the secondary subjects - these will anyway mostly be created via bulk upload, so hopefully not a big extra task for users. Event-centric data can usually be queried “the long way” or “via shortcuts” in RDF stores with inferencing. (So, I'm voting for solution b) - nested native modelling). Q7: What is meant with deterministic IRIs? Q8: Fundamental question - shouldn't we considering mapping as a connected rather than separate concern from the internal data model? This off-loads complexity from development to the user. |
|
I'm going to join the discussion with the first posted issues, and then try to come to Lozana's issues:
I'm neutral on this at this point. I would like to explore whether we could offer something from Pletka or if we could even produce as a derivative from Plekta an output which matched the neowiki specification. I have not explored LinkML but I know @sennierer is quite keen on it, would be good to hear more on it. Again maybe it is something Pletka could produce as a derivative. It would be good though that we produce or reuse a UI tool that can make the mapping and that the mapping is declarative and at least in principle readable to data modelling and design level users. The X3ML mapping standard that I mentioned in matrix thread is an example of this (without saying we should use it...)
I think the objective of having a round-trippable formalism is ideal BUT I'm not totally convinced that we know all the scenarios of how inverse transformation might work. Here is a scenario off the top of my head. We have a mapping from a NeoWiki scheme from a field that says birthdate and it takes date values and it has values like 01/01/1999. It is a rule in linked.art for example that a date value is always expressed as a date range. So the full proper transformation from Neowiki:Person/birthdate to Linked.Art would create both E21->p69i is born ->E65 Birth ->p4 has timespan -> E52 Timespan - P81a_ begin of the begin -> 00:01 | 01/01/199 and One bidirectional mapping would be the most elegant and efficient definitely. I feel like we might run into many edge cases where we essentially just have to write also an import mapping. Imagine also for example, in my NeoWiki:Person schema, I'm importing data from Europeana and let's say Europeana has a field for occupation and it allows as many as you want, but for some reason in my schema you can only have 5 (this is a silly example just want to give another edge case). Then I know the fields that match between Neowiki:Person and edm:Agent BUT in one direction I can have as many values as I want, in the other only up to five. edm:Agent is a good example as well because Europeana's Agent model covers both people and organizations. So it is easy to map out. Neowiki:Person creates an instance of edm:Agent. But if I wanted to pull in from edm:Agent then I need to do some sort of logic to only bring in persons amongst the agents and not organizations. This has to be specified going from Edm to Neowiki and not the other way, no?
Pletka doesn't create ontologies but allows you to take those ontologies and define schemas which are constructed following the chosen ontology's logic. How can Pletka interact with NeoWiki.
In the course of the ECHOLOT project we are going to document the main models to which we want to map in Pletka. This is important because it gives us one standard way to represent the models and to compare them. So there will be EDM, HDTO, Wikidata Profiles and ECHOLOT Models semantic projects in Pletka. This year on our roadmap is to make a semantic project to semantic project mapping tool. This is not made yet so we are very open to collaborating or otherwise exchanging ideas of what is needed here (or if this is not needed for ECHOLOT). So we could in principle then have a UI tool for mapping the ECHOLOT NeoWiki Models in a UI to the EDM Models and then exporting a derivative which is the format that Neowiki wants. To be discussed. Here is an example of how this looks like: https://pletka.io/projects/HDT So imagine that we also have this for EDM (actually we do but we are waiting for Europeana feedback to refine and release it), Wikidata profiles and ECHOLOT Models. Then we have a description of all the data standards in one consistent format. We can generate sparqls and shacls etc. for each, so this adds to our tooling for controlling the data outside of NeoWiki etc.
This is a great question, see my comments above.
I think this is a great conversation to have with T4.5 and T2.3.
I think the approach of one separate mapping per target schema (I wouldn't say ontology, ontologies aren't a target, they are just tools for building schemas) is the sensible solution. As is.
It kind of depends on the tool or tools we use to build them. It is very specialized work, so we can't assume that just anyone installing neowiki can or will want to map their system. Having the idea of packages though (see Arches project) is a really good idea. This could be aided by Pletka or done independently, but you would think you would want your neowiki instance to be able to install well built packages which contain both schemas (functioning and ready to go, for a certain kind of task) + the mappings (which is added value to export or import data into the system). The mappings should be understandable / readable / appproachable not just by technical experts but should have a form where the end user and at least a data modeller can go an understand from where they imported data or to where they are exporting data.
How does NeoWiki handle language generally? I haven't really clocked it.
I think the ECHOLOT project aims for multiple projections.
I don't know how we get away from some form of nesting. One needs to be able to bind multiple fields into a controllable unit: a name and its type and its language and its source as an example. All models need this.
|
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Discussion venue for
docs/planning/OntologyMapping.md— projecting native NeoWiki data into standard ontologies (CIDOC-CRM, EDM, …).Post here; substantive input from other channels (chats, calls, email) gets distilled into this thread with attribution. The doc remains the authoritative, updated statement of where the design stands — this thread is where it gets shaped. Closed when the doc graduates to an ADR.
All reactions