ML Wiki
Machine Learning Wiki - A collection of ML concepts, algorithms, and resources.

RDFS

semantic-web

RDFS

Is a schema language for RDF

  • roughly, RDF is for defining graphs, RDFS - for defining sets
  • RDFS tells how to use the graph structure - gives some semantics to the used vocabulary
    • how items and their properties are related
  • also provides some basic inferencing capabilities (for Knowledge Discovery)
    • Inference rules are nice to show with SPARQL CONSTRUCT queries
    • RDFS statements can be interpreted as FOL statements
    • for logical semantics behind there expressions see Semantic Web Logic
  • RDFS is expressed using RDF triples

RDFS “extends” RDF

  • in the sense that it gives some meaning to the triples

Basic Constructs

rdfs:Class

A set is identified in RDFS with rdfs:Class

  • note the use of rdfs: namespace
:AllStarPlayer rdf:type rdfs:Class. 
:MajorLeaguePlayer rdf:type rdfs:Class. 
:Surgeon rdf:type rdfs:Class. 
:Staff rdf:type rdfs:Class. 
:Physician rdf:type rdfs-subproperty:Class.

rdfs:subClassOf

Suppose we have the following assertions

  • :Apple rdfs:subClassOf :Fruit
  • :RedDelicions a :Apple
  • can infer that :RedDelicions a :Fruit

The inference rule is

CONSTRUCT { ?r rdf:type ?B } 
WHERE {
  ?A rdfs:subClassOf ?B .
  ?r rdf:type ?A
}

rdfs:subPropertyOf

Example:

  • relation “brother” is more specific than “sibling”
  • if smb is my brother, he also is my sibling
  • so :brother rdfs:subPropertyOf :sibling

The inference rule is ```text only CONSTRUCT { ?x ?r ?y } WHERE { ?x ?q ?y . ?q rdfs:subPropertyOf ?r }


Another example
- <img src="https://raw.githubusercontent.com/alexeygrigorev/wiki-figures/master/ufrt/xml/sw/rdfs-subproperty.png" alt="Image">


### <code>rdfs:domain</code> and <code>rdfs:range</code>
Typing data by usage (also - ''implicit'' typing)
- as opposed to the ''explicit'' typing <code>rdf:type</code>
- <code>rdfs:domain</code> - set of values for which a property is defined (subject)
- <code>rdfs:range</code> - set of values it can take (object)

Can define them as 
```carbon
CONSTRUCT {?y rdf:type ?D .} 
WHERE {
  ?P rdfs:range ?D . 
  ?x ?P ?y .
} 

CONSTRUCT {?x rdf:type ?D .} 
WHERE {
  ?P rdfs:domain ?D . 
  ?x ?P ?y .
}

’'’NB’’’:

  • there’s no notion of incorrect/inconsistent inference in RDFS
  • it doesn’t signalize an error if a property isn’t used consistently with the declaration
  • RDFS will infer the type to make this property consistent
  • this declaration is quite aggressive - even with one triple it can result in surprising inferences

Example

Suppose we have the following schema

  • :MarriedWoman rdfs:subClassOf :Woman.
  • :hasMaidenName rdfs:domain :MarriedWoman.
  • Image

And the following assertions:

  • :Karen :hasMaidenName "Stephens".

Inference:

  • Even if we don’t know that :Karen is a :Woman, we can infer that she’s married
  • infer that :Karen rdf:type :MarriedWoman based on rdfs:domain
  • infer that :Karen rdf:type :Woman based on rdfs:subClassOf

Modeling with RDFS

There are some basic modeling patterns that can be used in RDFS

Intersection

$C \equiv A \cap B$

  • :C rdfs:subClassOf :A
  • :C rdfs:subClassOf :B
  • need to be a subclass of both $A$ and $B$ to be in $C$

Union

$C \equiv A \cup B$

  • :A rdfs:subClassOf :C
  • :B rdfs:subClassOf :C
  • $x \in C$ when $x$ is a subclass of $A$, or subclass of $B$ (or both)

Properties

Intersection and Union can also be used for properties, e.g.

  • Intersection
    • :R rdfs:subPropertyOf :P .
    • :R rdfs:subPropertyOf :Q .
  • Union
    • :P rdfs:subPropertyOf :R .
    • :Q rdfs:subPropertyOf :R .

Example: ‘‘Terminology reconciliation’’

  • A military plane needs to determine if it can attach something or not
  • it has 2 sources of data
    • “never-target” list: schools, churches, hospitals
    • “off-limit airspace”: no-fly zones
  • a target is off-limit if it belongs to one of these classes
    • solution: use union
    • fc:Civilian rdfs:subClassOf cc:OffLimitTarget
    • space:NoFlyZone rdfs:subClassOf cc:OffLimitTarget

Example 2: Property union

  • if :A rdfs:label "something", then “something” is a printable name of :A
  • it’s more readable than URIs
  • but suppose that in our data source we don’t have it, but have something else with some textual information
    • :person1 :personName "James Dean"
    • :movie1 :movieTitle "Giant"
  • we want to use these properties as labels
  • solution: use property union:
    • :personName rdfs:subPropertyOf rdfs:label
    • :movieTitle rdfs:subPropertyOf rdfs:label

Non-modeling Properties in RDFS

There are properties that aren’t used for inference, but just for description

  • rdfs:label - text representation
  • rdfs:seeAlso - cross referencing
  • rdfs:isDefinedBy - primary source/description of a resource
  • rdfs:comment - a comment

See Also

Sources

✏️ Edit on GitHub