Repository navigation
Replies: 2 comments 1 reply
|
Until I read your post, I was convinced that the advantages of local properties outweigh the global ones, even though I come from the SMW world where only global properties exist. But I guess after having read your comment which I interpret as arguments for global properties, IMO the last paragraph summarizes it perfectly:
This is the main point. I guess NeoWiki can provide a great user experience using local properties. We don't have to have global properties in order to facilitate editing, for example. If I create a schema for person, I might be told that there is already a property called "name" in the organization schema and I might want to reuse for a very simple apporach or decide to create a first name and last name property instead (or in addition to a name property). And as much as I am in favor of supporting Semantic Web standards an methodologies, a lot of times users of NeoWiki will not care about it. They want to have a relational database model and maybe later want to be able to map their properties to some RDF vocabulary later. I encountered a lot of use cases like this. NeoWiki should support both useacases: where users want to adhere to a certain standard or re-use a certain ontology: there they want to have pre-defined and compliant properties. But sometimes they just want to define in their own therms what they want to manage and maybe map to vocabularies later. |
|
Thanks Dragan. This is exactly the kind of pushback we're after, and we've used it as input to our implementation work. You're right on several points. Properties carrying vetted meaning is real value. Unguided local modeling can get messy. Property suggestions are great. And your point stands: a single point of change needs a shared, referenced object. It's our read that you can get nearly all you want without global properties in the core model. To get the RDF you want, mappings are needed regardless, as seen in Wikibase. The choice between global and local properties affects what those mappings look like, while the RDF you get is the same. So NeoWiki does have a global layer in the mappings rather than via the core model properties. Those mappings are the shared, edit-once objects you describe: change how "Name" maps to EDM, and the projection of every Subject with that Schema follows. How mappings are best stored and edited, including whether all rules for one ontology live on a single page, is currently being worked out in #1065. With regards to standardization, guidance, and standard ontologies/RDF, what we have on master already puts NeoWiki ahead of Wikibase for this particular flow. Users can import bundles of Schemas and Mappings provided by ECHOLOT, so their museum (or whatever their use case is) gets all the schemas and properties it needs in NeoWiki and becomes accessible in various standard ontologies via the RDF API endpoints and the graph database(s). Schemas also improve over the Wikibase "for what properties should I add statements to this blank Item" situation, though there is room to go beyond what we currently have with those Notion-like suggestions you described. You can see the experimental mapping and RDF implementation live on the demo wiki at https://neowiki.dev/wiki/RDF_and_ontology_mappings, and perhaps by the time you read this, we'll have QLever running there so you can also run SPARQL queries. |
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
This is a response to the arguments for schema-local properties in Neowiki, as presented in the current planning document.
I want to apologize in advance if this contribution repeats points that have been stated elsewhere, I still hope it will be useful and only mildly annoying.
Starting with a bird's eye view, schema-local properties mean that Neowiki is reproducing a relational database: schemas are equivalent with tables containing rows (subjects) and columns (properties), each column having constraints on their input. Relations to entries in other tables can be created, but how the columns in the related table are semantically connected with the focus table, or the other way round, is to be defined per relation.
If I would want to use a relational database and project its content to RDF, I could use an existing open source project like nocodb, baserow, or teable (all of them with excellent UX), connect an existing, robust python RDF library with that tool's API, and map the columns per table to RDF terms.
Although this would only take moderate effort, I do not want to do this because I exactly want the semantics that the Semantic Web was designed to provide, and I do not want to potentially fumble it by having to manage these mappings for every schema (or in that case, "table") that I create, or manually enforce labeling rules, etc.
Going through the arguments against global properties one by one:
Technically it is true that RDF properties define an IRI, a data type, and sometimes, multi-language labels. What technical constraints apply to using a property is defined in schemas. However, a property represents much more than its technical definitions. It carries meaning that in cases of standardized ontologies is the result of years of research work, committee agreements, and, as a result, technical implementations in tools. For data curators, it is of high importance that they can, with confidence, pick a term of which they know what it means.
3 micro studies showing the opposite:
I would argue that high UX quality in this case would mean to guide users through such a change process, or designing it in a way that users cannot make common mistakes.
The smallest unit of information on the semantic web is a triple, not a property with a value.
City→Name→Brooklynis different fromPerson→Name→BrooklynorDog→Name→Brooklyn. In some cases, it makes sense to distinguish properties like in the city versus person example, but that depends on the goals users want to achieve with their data, how they want to query it, what comparisons and distinctions they deem productive. The tool they use should not prescribe such core things.That's right, this requires more development effort.
I'm not entirely sure about the exact Neowiki terms used here, but it looks like the solution to the problem is already described. Again, noted that it will require more effort.
As a summary of the two points above, a judgement can only be weighted against concrete features that would suffer from not getting the required attention if global properties were implemented instead. Without any insight into that, I would like to discuss the cost of not implementing global properties, which will mostly affect users, rather than developers, of Neowiki:
This is only true if you do not anticipate use beyond what a local, relational database could provide.
This is absolutely true. If Neowiki is implemented in a way that users do not realize that properties are not global, including, for instance, API calls that allow for reading and changing properties globally, the internal data model is of little relevance. It might just cause more unintended side effects and implementation effort down the line.
All reactions