Re: Feedback on RDF Graphs: Conceptual Role and Practical Use Cases

Hi Filip,

Thanks for asking the opinion of the community! Coincidentally I just 
blogged about how I pragmatically apply named graphs, even with left 
open semantics, in my projects: 
https://pietercolpaert.be/linkeddata/2025/09/30/named-graphs

The summary:

I use two distinct examples to set the scene:

  1. From the *RDF Stream Processing* CG: graphs hold statements valid 
in a specific time context (e.g., “My age = 36 on 2025-09-23”), letting 
consumers query only the context they care about.

  2. *UniProt*: graphs are just logical partitions (e.g., diseases, 
proteins) that are still globally true in the default graph.

The first is what I like to call a *contextual assertion*: the statement 
holds in a certain context. You can use the statements in your 
application if you explicitly agree to the context. We can see 
partitioning as a special case of this, as it makes an implicit 
assertion process for all applications. Also the case of quoting I see 
as a special case, where an application will probably never decide to 
assert the triples themself.

This way, even with the open semantics, tooling should always interpret 
named graphs as contextual assertions first, and should only with a 
manual intervention make it configurable to assert things from a named 
graph without more elaborate context selecting procedure in the default 
graph.

I also show an example from one of my projects, that is centered around 
Linked Data Event Streams (https://w3id.org/ldes/specification). There, 
we use named graphs there very much from the perspective of contextual 
assertions. However, we also bump into a limit of RDF1.1 (and 1.2) 
today, and it’s the fact that it’s difficult to “package” quads together 
to compile what I started calling “*RDF Messages*”. I kind of abuse 
named graphs for that now, but they should not have been necessary.  The 
post argues RDF needs a first-class notion of RDF messages: atomic units 
of quads with clear boundaries, similar to frames in Jelly RDF or 
similar to the idea of newline-delimited JSON-LD documents.

I also like the way that  Pierre-Antoine Champin is looking at it: to 
see RDF1.2 as an opportunity to do a conceptual mapping.

Kind regards,

Pieter