Re: RDF-star vs Wikidata for modelling Richard Burton

On 12/8/21 03:07, Pierre-Antoine Champin wrote:

>
> On 07/12/2021 22:25, Peter Patel-Schneider wrote:
>> On Tue, 2021-12-07 at 14:52 +0100, Pierre-Antoine Champin wrote:
>>> In LPGs as well, there is a point were modelling marriages or pipes
>>> as edges will hit a wall. More specifically, if you want to relate
>>> them to other *nodes* of the graph (link a marriage to its location,
>>> link a pipe to its manufacturer), you also need to reifiy them as
>>> nodes.
>> I see some formulations of property graphs where edges can only be
>> related to values, but is that really the case for all property graphs?
>
> I have never seen an explicit definition of "Property Graphs" where this was 
> not the case (but I have not seen them all, of course).
>
> I also have the feeling that allowing nodes as property values would pervert 
> the idea of PGs, where the distinction between the internal structure of 
> nodes (properties) and relation with other nodes (edges) is considered an 
> important feature.
>
>> (Well, if you think of Wikibase as a property graph formulation then
>> there is at least one counterexample.)
>
> It never occurred to me to put Wikibase in the "Property Graph" family. I 
> see some similarities, but also many differences. In particular, the 
> distinction I mentioned above (between "properties" and "edges" of a node, 
> in PG parlance) does not exist in Wikibase, where there all represented as 
> "statements" (except for labels and descriptions... but I hope you see my 
> point).
>
>   pa


If by property graphs you mean only the stuff that fits into the data model 
underlying the Property Graph Schema Working Group, then edges can only be 
related to values.  But this limitation is not present in several large 
efforts that deserve to be called labelled property graphs, notably RDF and 
Wikibase, both of which are based on labelled property graphs of one flavour 
or another.

Every time I look at limited property graphs, where only attribute-value pairs 
can be associated with edges, I find that my use case doesn't fit.  For 
provenance, I want to attach property-object information to edges.  In other 
cases, I want multi-valued attributes.  In yet other cases, I want 
sophisticated structures for values and thus the relationship has to be a 
property and can't be attached to edges.  So if I wanted to use the property 
graph system I would have to reify many relationships, so there is no reason 
to move from RDF to limited property graphs.

I understand that the property graph people want to have a rough standard.  
But why the limitations, which as far as I can tell, come from some old 
storage or speed optimization.


peter



peter

Received on Thursday, 9 December 2021 16:36:27 UTC