Re: RDF-star use cases from Amazon Neptune

> Am 05.12.2021 um 20:08 schrieb Pierre-Antoine Champin <pierre-antoine.champin@ercim.eu>:
> 
> 
> On 05/12/2021 02:46, Peter F. Patel-Schneider wrote:
>> Although RDF is in some sense a logic, this is not really relevant to the discussion here.
> I still beg to differ. See below
>> 
>> There is no requirement that a logic be based on sets (or set-like graphs) instead of multi-sets (or multi-set-like graphs).
> 
> That's right, but RDF as a logic is.
> 
> Rephrasing my point below: this "restriction" of RDF is not an arbitrary choice that could be easily revised, because it is linked to the underlying semantics of RDF. Furthermore, since PGs do not have an underlying semantics, the same construct can be used with a different meaning in different contexts (e.g. some edges are considered as automatically asserted, while other edges are only asserted conditionally to some properties). So even if we changed the RDF abstract syntax to better align with the PG data model, I expect that there would still be common PG patterns that would not map well to the new RDF's semantics.

One of my criticisms concerning the proposed semantics is that it makes it hard to model even the most central modelling primitive of LPG, the very essence of the LPG model: a primary relation annotated with secondary attributes.

Property graphs don’t have formal semantics but of course they have semantics, as everything does, and these semantics are pretty strong and expressive. LPGs partition the knowledge space in two spheres: the sphere of primary objects and primary relations between those primary objets, and the sphere of secondary attributes on both objects and relations. In a setting where an application sets a focus (of what is primary) this makes the knowledge space much more manageable than the flat structure of RDF where every relation is of equal importance (but where OTOH every relation can be primary, depending on application needs).

It comes with the LPG model that just about any thinkable relation can be the primary relation in some application scenario: "married", "buys", "likes", "inherts", "belongsTo", "staysAt" etc etc. That fueled my criticism of the TEP proposal which IMO is driven by the perception that it will be needed rarely in practice. Quite to the contrary I think that if you seriously consider using RDF-star embedded triples to represent LPG-style primary relations (to be annotated with secondary attributes) then you will see that just about any property defined so far on the semantic web will have to be re-defined as a TEP-property. This concern is of course only valid if you want reasoning - but if you don’t want reasoning, then why use RDF(-star)?

That vague reference that there will always be corner cases - as justified as it always is - shouldn't obscure the issues that the specific semantics defined in the RDF-star proposal has with the very essence of LPG semantics. If the embedded triple was defined as being referentially transparent that problem would disappear. If another syntax was defined as being referentially transparent - like a statement identifier, or the shorthand syntax - that problem would probably be manageable. But as it stands RDF-star will either be prohbitively cumbersome to represent LPGs, making RDF-star useless for this use case, or the embedded triple will be used as a syntax but the defined semantics will be ignored, rendering the RDF-star semantics as defined useless.

Thomas


>> SPARQL and SPARQL* do not use the logic of RDF.  They have no more semantic commitment than retrieval from property graphs.
> 
> Granted. People can use RDF while totally ignoring its semantics, and still query it with SPARQL and maybe get something useful from it. But would that still be RDF? If they published this data on the web, would it be desirable that they advertise it as RDF?
> 
>   pa
> 
>> 
>> 
>> peter
>> 
>> 
>> 
>>> On 12/3/21 6:31 AM, Pierre-Antoine Champin wrote:
>>>> I think that presenting this feature of RDF as a "restriction" is unfair, and misses the point. In my view, the impedance mismatch between RDF and PGs is not due to some arbitrary restriction on the RDF model. It is due to the fact that RDF is a logic, that can be represented as a graph, while PG is a graph data model, without any semantic commitment.
>> 
> <OpenPGP_0x9D1EDAEEEF98D438.asc>