Looking for a SHACL Shape equivalent

[Scott Henninger]

Another SHACL question for the topbraid-users group:

This one is a bit hard to explain, but the general idea is to create a shape that requires two target classes, and the hierarchical parent must be one of three classes, and the target cannot have any properties are a sub-property of a property.

Working it out a bit I believe the following is the equivalent query in SPARQL:

SELECT * 

WHERE {

   $this a ex:Cls1, ex:Cls2  .

   { $this ex:parent/rdf:type ex:Cls3 . }

   UNION

   { $this ex:parent/rdf:type ex:Cls4 . }

   UNION

   { $this ex:parent/rdf:type ex:Cls5. }

   ?prop rdfs:subPropertyOf ex:assoc .

   $this ?prop ?x

}

Any $this found by this query would be a violation, based on the last triple pattern - i.e. the rest are requirements,that are taken up by other shapes.  But what I want here is to report if the target class combination, with appropriate parents, has one of the ex:assoc sub-properties.  To paraphrase: "Instances of Cls1 and Cls2 ... cannot have any ex:assoc properties.

Any hints or direction on this would be useful.  I've tried a few things but have so far failed to make much headway.

Appreciate it and TIA.

-- Scott

[Holger Knublauch]

Hi Scott,

I may not fully understand your scenario, but I don't think this can be expressed in SHACL Core. The problem is not so much in the matching pattern (with the types etc) but in the subPropertyOf link. The last line would need to become something with sh:qualifiedMinCount 1 but you would need a sh:path and the sh:path cannot be computed dynamically, i.e. you cannot use a variable that gets computed using subPropertyOf. It would only work if you can know the sub-properties in advance, when you write the constraint.

You have it formulated in SPARQL already, so why not stay in SHACL-SPARQL? Does your target platform only support SHACL Core?

Holger

--
You received this message because you are subscribed to the Google Groups "TopBraid Suite Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to topbraid-user...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/topbraid-users/3a55f966-d99d-4d32-875e-3e32a1f643c4n%40googlegroups.com.

[Scott Henninger]

Unfortunately we do not have SPARQL support for SHACL, as we use an old version of Semaphore. Maybe someday in the future.

In the meantime, what about if we drop the sub-property triple pattern and have a known set of (or single) properties? Can that be expressed in SHACL? Particularly the part about targeting multiple classes and multiple parent classes.

Thanks, Holger! I appreciate the help.

-- Scott

On Sep 3, 2021, at 8:06 PM, Holger Knublauch <hol...@topquadrant.com> wrote:



To view this discussion on the web visit https://groups.google.com/d/msgid/topbraid-users/cca816bc-26db-4af4-2881-4ee80ff7866e%40topquadrant.com.

[Holger Knublauch]

On 2021-09-05 3:38 am, Scott Henninger wrote:

Unfortunately we do not have SPARQL support for SHACL, as we use an old version of Semaphore. Maybe someday in the future.

In the meantime, what about if we drop the sub-property triple pattern and have a known set of (or single) properties? Can that be expressed in SHACL? Particularly the part about targeting multiple classes and multiple parent classes.

Ok, if the properties are known you can use a pattern such as

    sh:path [ sh:alternativePath ( ex:prop1 ex:prop2 ex:prop3 ) ] ;  # ex:prop1|ex:prop2|ex:prop3

The rest could roughly be

ex:Shape
    a sh:NodeShape ;
    sh:targetClass ex:Cls1 ;
    sh:not [
        sh:class ex:Cls2 ;
        sh:property [
            sh:path ex:parent ;
            sh:qualifiedValueShape [
                sh:or ( [ sh:class ex:Cls3 ] [ sh:class ex:Cls4 ] [ sh:class ex:Cls5 ] ) ;
            ] ;
            sh:qualifiedMinCount 1 ;
        ) ;
        sh:path [ sh:alternativePath ( ex:prop1 ex:prop2 ex:prop3 ) ] ;  # ex:prop1|ex:prop2|ex:prop3
        sh:minCount 1 ;
    ]

(above I assume your rdf:type matches can be expressed using sh:class, i.e. subclasses also count).

I am not sure if the above represents your intention, but the mechanism that I have used was basically to match the pattern that you have in the WHERE clause and put a sh:not around it, meaning that a violation will be reported if the instance of Cls1 also matches the given pattern.

Holger

To view this discussion on the web visit https://groups.google.com/d/msgid/topbraid-users/49677E3C-2B4F-488F-BB6B-90881B496020%40gmail.com.

[Scott Henninger]

Thanks for the help Holger, it gave me some ideas, which unfortunately haven't panned out yet.  First, I take it that the parent in the third-to-last line of the example should be a square bracket.  Or there's something else missing in the syntax.

The most confounding part to me is how to express the class constraints, so let me focus there with three example data sets.

This is a valid shape.  The target class has both Cls1 and Cls2 and the parent class has one of Cls{3, 4, 5}:

   ex:target1
       a ex:Cls1, ex:Cls2 ;
       ex:parent ex:target2 .

   ex:target2
       a ex:Cls3 .

This is an invalid shape only because of the extra class definition in target1.  The extra class definition in target2 is OK.

   ex:target1
       a ex:Cls1, ex:Cls2, ex:Cls6;
       ex:parent ex:target2 .

   ex:target2
       a ex:Cls3, Cls7 .

This is also an invalid shape because target2 does not have a class definition in Cls{3, 4, 5}:

   ex:target1
       a ex:Cls1, ex:Cls2;
       ex:parent ex:target2 .

   ex:target2
       a ex:Cls7 .

Hopefully this helps some.  As arbitrary as it seems, the examples come from real requirements, and while I'm pretty familiar with targeting a class, in this case two classes are being targeted along with a property-to-type relationship. That I'm unclear on, so any hints on any pieces of this puzzle would be useful.  

Thanks again for your efforts!

-- Scott

To view this discussion on the web visit https://groups.google.com/d/msgid/topbraid-users/baad61dd-3f49-84e6-09d1-8db48331c1af%40topquadrant.com.

[Holger Knublauch]

On 2021-09-07 3:07 pm, Scott Henninger wrote:

Thanks for the help Holger, it gave me some ideas, which unfortunately haven't panned out yet.  First, I take it that the parent in the third-to-last line of the example should be a square bracket.  Or there's something else missing in the syntax.

Right, it was lacking a sh:property [ ... ] around the last two rows. It's hard to help on such issues without real test data, so moving forward it would be best to get real instances, real class definitions and a description of invalid instances so that we only need to fill in the blanks. I see you have started this below.

The most confounding part to me is how to express the class constraints, so let me focus there with three example data sets.

This is a valid shape.  The target class has both Cls1 and Cls2 and the parent class has one of Cls{3, 4, 5}:

   ex:target1
       a ex:Cls1, ex:Cls2 ;
       ex:parent ex:target2 .

   ex:target2
       a ex:Cls3 .

This is an invalid shape only because of the extra class definition in target1.  The extra class definition in target2 is OK.

   ex:target1
       a ex:Cls1, ex:Cls2, ex:Cls6;
       ex:parent ex:target2 .

   ex:target2
       a ex:Cls3, Cls7 .

This case could become

ex:Shape
    a sh:NodeShape ;

    sh:targetClass ex:Cls1, ex:Cls2 ;
    sh:not [ sh:class ex:Cls6 ] ;

to state that instances of Cls1 or Cls2 cannot also be instances of Cls6. It does NOT state that they need to have both types 1 and 2.

This is also an invalid shape because target2 does not have a class definition in Cls{3, 4, 5}:

   ex:target1
       a ex:Cls1, ex:Cls2;
       ex:parent ex:target2 .

   ex:target2
       a ex:Cls7 .

This case could become

ex:Shape
    a sh:NodeShape ;

    sh:targetClass ex:Cls1, ex:Cls2 ;

    sh:property [
        sh:path ex:parent ;

        sh:or (
            [ sh:class ex:Cls3 ]
            [ sh:class ex:Cls4 ]
            [ sh:class ex:Cls5 ]
       )

   ]

But this logic feels different from what you had in your original SPARQL query, which seemed to be more "exists" than "for-all".

Note that for general non-product related questions you may get better breadth of input on places like StackOverflow or the https://www.w3.org/community/shacl/ mailing list or similar places. This is not to discourage you from asking here, but to understand that we cannot give such requests the same amount of attention as customer questions, so my input may be rather cursory than in-depth.

Thanks,
Holger

To view this discussion on the web visit https://groups.google.com/d/msgid/topbraid-users/CAJ-KvVqCG_3HZUDM-NL_PR4i%2B6c4RQfL%3DyNjR5Odco5exs%3DJMg%40mail.gmail.com.