Lab: RDFS: Difference between revisions

From info216
No edit summary
mNo edit summary
 
(15 intermediate revisions by 2 users not shown)
Line 1: Line 1:
=Lab 6: RDFS Programming with rdflib and owlrl=
==Topics==
==Topics==
Basic RDFS graph programming in RDFlib.
* Simple RDFS statements/triples
Entailments and axioms with owlrl.
* Basic RDFS programming in RDFlib
 
* Basic RDFS reasoning with OWL-RL
==Classes/Methods/Vocabularies==
owlrl.RDFSClosure (RDFS_Semantics, closure, flush_stored_triples)


'''Vocabularies: '''
==Useful materials==
rdflib classes/interfaces and attributes:
* RDF (RDF.type)
* RDFS (RDFS.domain, RDFS.range, RDFS.subClassOf, RDFS.subPropertyOf)
* [https://docs.google.com/presentation/d/13fkzg7eM2pnKGqYlKpPMFIJwnOLKBkbT0A62s7OcnOs Lab Presentation of RDFS]


RDF.type
OWL-RL:
* [https://pypi.org/project/owlrl/ OWL-RL at PyPi]
* [https://owl-rl.readthedocs.io/en/latest/ OWL-RL Documentation]


RDFS.subClassOf, RDFS.subPropertyOf, RDFS.domain, RDFS.range, RDFS.label, RDFS.comment,  
OWL-RL classes/interfaces:
* RDFSClosure, RDFS_Semantics


==Tasks==
==Tasks==
First, pip install owlrl.
'''Task:'''  
The RDFS Vocabulary can be imported from rdflib.namespace, just like FOAF or RDF.
Install OWL-RL into your virtual environment:
 
pip install owlrl
'''Consider the following Scenario:'''
"University of California and University of Valencia are both Universities.
All universities are higher education institutions (HEIs). Only persons can have an expertise, and what they have expertise in is always a subject. Only persons can graduate from a HEI. If you are a student, you are in fact a person as well. That a person is married to someone, means that they know them. Finally, if a person has a name, that name is also the label of that entity."
 
'''Create RDFS triples corresponding to the text above with RDFlib''' - if you can, try to build on
your example from lab 2!
 
To create the graph in python, you can just use the g.add syntax as we have done previously, or you can use the following code sample to parse a file into a graph:


<syntaxhighlight>
'''Task:'''
from rdflib import Graph, Namespace
We will use simple RDF statements from the Mueller investigation RDF graph you create in Exercise 1. Create a new rdflib graph and add triples to represent that:
import owlrl
* Rick Gates was charged with money laundering and tax evasion.


Use RDFS terms to add these rules as triples:
* When one thing that is charged with another thing,
** the first thing is a person under investigation and
** the second thing is an offense.


# Create the graph
To add triples, you can use either:
g = Graph()
* simple ''graph.add((s, p, o))'' statements or
* ''INSERT DATA {...}'' SPARQL updates.
If you use SPARQL updates, you can define a namespace dictionary like this:
EX = Namespace('http://example.org#')
NS = {
    'ex': EX,
    'rdf': RDF,
    'rdfs': RDFS,
    'foaf': FOAF,
}
You can then give NS as an optional argument to graph.update() - or to graph.query() - like this:
g.update("""
    # when you provide an initNs-argument, you do not have
    # to define PREFIX-es as part of the update (or query)
    INSERT DATA {
        # the triples you want to add go here,
        # you can use the prefixes defined in the NS-dict
    }
""", initNs=NS)


# Parse input data into the graph, format is dependent on the file format. Here turtle (ttl). And location is the path to the local file
'''Task:'''
g.parse(location="input.ttl", format="turtle")
* Write a SPARQL query that checks the RDF type(s) of Rick Gates in your RDF graph.
</syntaxhighlight>
* Write a similar SPARQL query that checks the RDF type(s) of money laundering in your RDF graph.
* Write a small function that computes the ''RDFS closure'' on your graph.
* Re-run the SPARQL queries to check the types of Rick Gates and of money laundering again: have they changed?


You can compute the RDFS closure on a graph ''g'' like this:
import owlrl
owlrl.DeductiveClosure(owlrl.RDFS_Semantics).expand(g)


Using these three lines we can add automatically the inferred triples (like ex:University rdf:type ex:Higher_Education_Institute) :
'''Task:'''
<syntaxhighlight>
Use RDFS terms to add this rule as a triple:
rdfs = owlrl.RDFSClosure.RDFS_Semantics(g, False, False, False)
* A person under investigation is a FOAF person.
rdfs.closure()
* Like earlier, check the RDF types of Rick Gates before and after running RDFS reasoning. Do they change?
rdfs.flush_stored_triples()
</syntaxhighlight>


After you have done this, try to add the following scenario to you graph as well:
'''Task:'''
"Having a degree from a HEI means that you have also graduated from that HEI. That a city is a capital of a country means that this city is located in that country. That someone was involved in a meeting, means that they have met the other participants. If someone partook in a meeting somewhere, means that they have visited that place"
Add in "plain RDF" as in Exercise 1:
To do this, you will have to swap out the line
* Paul Manafort was convicted for tax evasion.
<syntaxhighlight>
rdfs = owlrl.RDFSClosure.RDFS_Semantics(g, False, False, False)
</syntaxhighlight>
with
<syntaxhighlight>
rdfs = owlrl.OWLRL.OWLRL_Semantics(g, False, False, False)
</syntaxhighlight>
As some of these triples require more advanced reasoning.


Check that simple inference works - make sure that your graph contains triples like these, even if
Use RDFS terms to add these rules as triples:
you have not asserted them explicitly:
* When one thing is ''convicted for'' another thing,
* that University of California and Valencia are HEIs
** the first thing is also ''charged with'' the second thing.
* that Cade, Emma, and Mary are all persons
* that Cade and Emma have both graduated from some HEI
* that Cade knows Mary


One way to check if the triples are there:
''Note:'' we are dealing with a "timeless" graph here, that represents facts that has held at "some points in time", but not necessarily at the same time.
<syntaxhighlight>
universities = g.query("""
PREFIX ex: <http://example.org/>
ASK {
    ex:University_of_California rdf:type ex:Higher_Education_Institution.
}
""")
print(bool(universities))
</syntaxhighlight>


Rewrite some of your existing code to use rdfs:label in a triple and add an rdfs:comment to the same resource.
* What are the RDF types of Paul Manafort and of tax evasion before and after RDFS reasoning?
* Does the RDFS domain and range of the ''convicted for'' property change?


==If you have more time...==
==If you have more time...==
Create a new RDFS graph that wraps an empty graph. This graph contains only RDFS axioms. Write it out in Turtle and check that you understand  the meaning and purpose of each axiom.
'''Task:'''
 
* Create a Turtle file with all the RDF and RDFS triples from the earlier tasks.  
Create an RDF (not RDFS) graph that contains all the triples in your first graph (the one with all the people and universities). Subtract all the triples in the axiom graph from the people/university graph. Write it out to see that you are left with only the asserted and entailed triples and that none of the axioms remain.
* Go to a local or online Blazegraph. Create a new Blazegraph namespace and ''Use'' it. (Remember that namespaces in Blazegraph and rdflib have nothing to do with one another.)  
 
* Go to the ''Upload'' tab to add the triples into your new Blazegraph namespace.
<!-- Download the SKOS vocabulary from https://www.w3.org/2009/08/skos-reference/skos.rdf and save it to a file called, e.g., SKOS.rdf .
* Go through each of the above queries to confirm that Blazegraph has not performed any RDFS reasoning.
Use the schemagen tool (it is inside your Jena folders, for example under apache-jena-3.1.1/bin) to generate a Java class for the SKOS vocabulary.  
You need to do this from a console window, using a command like "<path>/schemagen -i <infile.rdf> -o <outfile.java>".


Copy the SKOS.java file into your project in the same package as your other Java files,  and try to use SKOS properties
You can also list all the triples in the namespace to ensure that nothing has been added:
where they fit, for example to organise the keywords for interests and expertise.
SELECT * WHERE { ?s ?p ?o }
-->


==Useful Readings==
'''Task:'''
*[https://wiki.uib.no/info216/index.php/File:S05-RDFS-11.pdf Lecture Notes]
* Create another Blazegraph namespace, but this time ''with the inference box checked''. ''Use'' the new namespace.  
*[https://wiki.uib.no/info216/index.php/Python_Examples Example page]
* List all the triples in the new inferencing namespace. (What you see are the axioms in RDFS.) How many are there?
* Upload the triples into your new inferencing Blazegraph namespace.
* List all the triples again. How many have been added?
* Go through each of the above queries to ensure that Blazegraph (with inference) has reasoned as you expect.

Latest revision as of 12:14, 15 March 2023

Topics

  • Simple RDFS statements/triples
  • Basic RDFS programming in RDFlib
  • Basic RDFS reasoning with OWL-RL

Useful materials

rdflib classes/interfaces and attributes:

OWL-RL:

OWL-RL classes/interfaces:

  • RDFSClosure, RDFS_Semantics

Tasks

Task: Install OWL-RL into your virtual environment:

pip install owlrl

Task: We will use simple RDF statements from the Mueller investigation RDF graph you create in Exercise 1. Create a new rdflib graph and add triples to represent that:

  • Rick Gates was charged with money laundering and tax evasion.

Use RDFS terms to add these rules as triples:

  • When one thing that is charged with another thing,
    • the first thing is a person under investigation and
    • the second thing is an offense.

To add triples, you can use either:

  • simple graph.add((s, p, o)) statements or
  • INSERT DATA {...} SPARQL updates.

If you use SPARQL updates, you can define a namespace dictionary like this:

EX = Namespace('http://example.org#')
NS = {
    'ex': EX,
    'rdf': RDF,
    'rdfs': RDFS,
    'foaf': FOAF,
}

You can then give NS as an optional argument to graph.update() - or to graph.query() - like this:

g.update("""
    # when you provide an initNs-argument, you do not have 
    # to define PREFIX-es as part of the update (or query)

    INSERT DATA {
        # the triples you want to add go here,
        # you can use the prefixes defined in the NS-dict
    }
""", initNs=NS)

Task:

  • Write a SPARQL query that checks the RDF type(s) of Rick Gates in your RDF graph.
  • Write a similar SPARQL query that checks the RDF type(s) of money laundering in your RDF graph.
  • Write a small function that computes the RDFS closure on your graph.
  • Re-run the SPARQL queries to check the types of Rick Gates and of money laundering again: have they changed?

You can compute the RDFS closure on a graph g like this:

import owlrl

owlrl.DeductiveClosure(owlrl.RDFS_Semantics).expand(g)

Task: Use RDFS terms to add this rule as a triple:

  • A person under investigation is a FOAF person.
  • Like earlier, check the RDF types of Rick Gates before and after running RDFS reasoning. Do they change?

Task: Add in "plain RDF" as in Exercise 1:

  • Paul Manafort was convicted for tax evasion.

Use RDFS terms to add these rules as triples:

  • When one thing is convicted for another thing,
    • the first thing is also charged with the second thing.

Note: we are dealing with a "timeless" graph here, that represents facts that has held at "some points in time", but not necessarily at the same time.

  • What are the RDF types of Paul Manafort and of tax evasion before and after RDFS reasoning?
  • Does the RDFS domain and range of the convicted for property change?

If you have more time...

Task:

  • Create a Turtle file with all the RDF and RDFS triples from the earlier tasks.
  • Go to a local or online Blazegraph. Create a new Blazegraph namespace and Use it. (Remember that namespaces in Blazegraph and rdflib have nothing to do with one another.)
  • Go to the Upload tab to add the triples into your new Blazegraph namespace.
  • Go through each of the above queries to confirm that Blazegraph has not performed any RDFS reasoning.

You can also list all the triples in the namespace to ensure that nothing has been added:

SELECT * WHERE { ?s ?p ?o }

Task:

  • Create another Blazegraph namespace, but this time with the inference box checked. Use the new namespace.
  • List all the triples in the new inferencing namespace. (What you see are the axioms in RDFS.) How many are there?
  • Upload the triples into your new inferencing Blazegraph namespace.
  • List all the triples again. How many have been added?
  • Go through each of the above queries to ensure that Blazegraph (with inference) has reasoned as you expect.