Backwards chaining (also known as query rewriting) is an approach to semantic reasoning where queries are rewritten based on rules.
For instance, we might have this data:
:alice a :Woman.
:bob a :Man .
:sam a :Person .
:Woman rdfs:SubClassOf :Person .
:Man rdfs:SubClassOf :Person .
Querying for all members of the :Personclass, will look like this:
?person a :Person
Now, this pattern will only match Sam, as neither Alice nor Bob directly belong to the :Person class.
So a reasoner that uses backward chaining will rewrite the query to something like this:
?person a/rdfs:subClassOf* :Person
Where we use a property path to say that we are interested in paths beginning with a, and having 0 or more instances of rdfs:subClassOf.
If we now add these two triples:
:Charlie a :Boy .
:Boy rdfs:subClassOf :Man .
We will add further depth to our class hierarchy.
But again, rewriting the query will work.
Backwards Chaining vs Forwards Chaining (aka materialisation)
Backwards Chaining Advantages
-Can be used in conjunction with virtual graphs
-Does not need to do any upfront computation when ingesting a rule
Forward Chaining Disadvantages
-Reasoning cannot be used to improve performance
-In the past, reasoning was slow and has garnered a now unjust reputation for poor performance