Inferencing in Semantic Web

In Semantic Web, using RDFS and OWL many things can be inferred based on facts that are stored in the RDF triple store


Inferencing - a systematic process of adding new tuples to an RDF graph based on some patterns (rules)

  • asserted triples - RDF triples provided by some data source
  • inferred triples - new triples added by inference rules
  • inference rules - systematic patterns that define how and what to infer
  • inference engine - engine that does the inference


Motivating Example

Suppose you have a SPARQL query on your RDF graph

  • you look for :RedDelicious apples
  • you are doing it in the :Fruit section
  • but the result is empty, because :RedDelicious is an :Apple, not a :Fruit
    • i.e. :RedDelicious a :Apple
    • and :Apple :subClassOf :Fruit

Possible solution:

SELECT ?item 
WHERE {
  ?class :subClassOf* :Fruit . 
  ?item a ?class . 
} 


But users will have to do it each time

  • alternatives?
  • can have rules if $X$ subclass of $Y$, then $\forall x \in X: x \in Y$
  • inferencing - given some information we can determine related information - and consider that it's also stored in our database
  • so here we'd infer that if :RedDelicious is an :Apple, it's also a :Fruit
  • :Fruit is broader than :Apple, so :Fruit is a subclass of :Apple


Inferencing

The motivating example illustrates the type propagation rule

  • this is a part of the RDFS language: rdfs:subClassOf relation
  • rule: $X$ rdfs:subClassOf $Y \Rightarrow $ every member of $X$ is also a member of $Y$

Reference rules of RDFS/OWL can be expressed using SPARQL Construct queries:

  • this is a good way of describing rules
CONSTRUCT { ?r rdf:type ?B } 
WHERE {
  ?A rdfs:subClassOf ?B 
  ?r rdf:type ?A
} 


Implementation Details

When the inference happens?

  • Cached Inference:
    • inferred triples are stored along with asserted
    • risk an explosion of the triple store
    • also, change management is important - how to propagate changes and deletes
    • for deletes - same inferred tuple can be due to several facts, so need to be careful when deleting
  • Just-In-Time Inference
    • To respond to queries only
    • no inferred triples retained


Sources