Basic semantic graph transformations

This page is following Basic graph transformations but with a semantic graph perspective.

Rich/poor level of information

A single triple is a "poor piece of information", but other triples with the same subjects can build a rich set of information.

Poor piece of information: s p o .

Rich piece of information:

s p o ;
  a S .
o a O .

Time management

Rich piece of information: adding time information:

s p_23DEC2018 o ;
  a S .
o a O .
p_23DEC2018 a P .
P a Time_Predicate .

This is practical because s P o . can be deduced easily even if s p_23DEC2018 o . is more precise.

The statement P a Time-Predicate . indicates that P is a time-enabled predicate.

Version management

We can have a variation of what we saw with version tagging.

s p_V2 o ;
  a S .
o a O .
p_V2 a P .
P a Version_Predicate .

Managing life-cycle

Case of subject modification and history keeping.

Step 1. We have:

s1 p o .

Step 2: s1 becomes s2. We have:

s1 p o .
s2 previous s1 .
s2 p o . // "Rewiring"

This is ambiguous because the 3rd statement was made after the first. Let's use a time predicate.

s1 p_12DEC2018 o .
p_12DEC2018 a P .
P a Time_Predicate .
s2 previous s1 .
s2 p_23DEC2018 o . // "Rewiring"
p_23DEC2018 a P .

We can look at the graph at various moments.

Indeed, we still have:

s1 P o .
s2 previous s1 .
s2 P o . // "Rewiring"

But we also encoded a more precise information.

We could think about removing s1 p o . after s2 is created but as the semantic web is a cumulative system, this does not seem very interesting.

Shortcuts

Let's consider the pattern:

q = p(1) o p(2) o ... o p(n)

with p(i) a set of predicates.

s q a .
if
x(n) p(n) a .
and x(n-1) p(n-1) x(n) .
and ..
and s p(1) x(2) . 

q id just a new "predicate name".

This can be very useful to present the same reality in another perspective/point of view.

Filters

If we have: s p o ; q a .

We can define a subgraph by "removing" the q predicate:

graph(s , depth=1) \ {q} => s p o .

Classical inferences

The use of classical inferences is very important also.

Temporal inferences

If we have:

s p(t1) a .
s p(t2) b .
p(t1) a P .
p(t2) a P .
P a Time_Predicate .

and a and b were not existing before the predicates were created, we could deduce:

a before b .

Not sure it is useful.

Simpler like that:

a2 previous a1 .
a3 previous a2 .
=> a3 previous a1 .

Some predicates can have special transitivity features.

Special predicates

  • Transitive predicates: a r b . and b r c . => a r c .
  • Commutative predicates: a r b . => b r a .
  • Semantic inversion (very used in Owl): p is the inverse of r, so if a r b . => b p a .

To do

  • Re-analyze the list notion in the RDF standard
  • Re-analyze the Gruf temporal display
  • Formalization of grammar in table parsing => grammar being a semantic graph
  • List graph transformations in Sparql-like