> On 2 Jan 2025, at 8:46 AM, Vladimir Alexiev <vladimir.alexiev@graphwise.ai> wrote: > > One motivation to use a dedicated datalog-like RDF-based inference "language" > is truth maintenance: to be able to trace which triples were or may have been inferred from which other triples, > so you can retract inferred triples upon deletion of explicit triples. I agree on this requirement. In our experience, it is often sufficient to compute certain "derived" property values on demand, based on sh:values. > > That is how GraphDB "pie" rules work (https://graphdb.ontotext.com/documentation/10.8/reasoning.html). > They have various restrictions (no arithmetics, comparisons or negation) that allow such "smooth deletion (https://graphdb.ontotext.com/documentation/10.8/delete-optimisations.html). > In contrast, I am not aware of any work on traceability of triples derived with SPARQL. > Even if you represent SPARQL in RDF, I think the traceability task is very hard or even impossible. > <favicon.ico> Maybe these rules could be defined as a subset of SPARQL CONSTRUCT syntax? With an RDF representation, a profile could be defined for various subsets. Holger
Received on Thursday, 2 January 2025 08:09:00 UTC