Pattern Miner Prospective Examples

From OpenCog
Jump to: navigation, search

This page gathers together some examples of patterns that we would like to have mined by the Pattern Miner in the relatively near term (written May 1, 2017):

patterns in English and Lojban NLP parses

(Ruiting)

patterns in PLN inference histories

Context-free patterns

An example here would be "basic chaining of deductions"

ExecutionOutputLink
  GroundedSchemaNode "deduction-formula"
  ListLink
    ImplicationLink
      VariableNode "$A"
      VariableNode "$C"
    ExecutionOutputLink
      GroundedSchemaNode "deduction-formula"
      ListLink
        ImplicationLink
          VariableNode "$A"
          VariableNode "$B"
        ImplicationLink
          VariableNode "$A"
          VariableNode "$B1"
        ImplicationLink
          VariableNode "$B1"
          VariableNode "$B"
    ExecutionOutputLink
      GroundedSchemaNode "deduction-formula"
      ListLink
        ImplicationLink
          VariableNode "$B"
          VariableNode "$C"
        ImplicationLink
          VariableNode "$B"
          VariableNode "$B2"
        ImplicationLink
          VariableNode "$B2"
          VariableNode "$C"

This says

  • A implies B1, and B1 implies B, therefore A implies B
  • B implies B2, and B2 implies C, therefore B implies C
  • A implies B, and B implies C, therefore A implies C

which can be graphically represented by the upside-down inference tree

Implication A B1  Implication B1 B   Implication B B2   Implication B2 C
--------deduction-formula---------   ---------deduction-formula---------
         Implication A B                       Implication B C
         -----------------deduction-formula-------------------
                           Implication A C

Context-dependent patterns

An example would be

InheritanceLink
  VariableNode "$P"
  ConceptNode "Person"

Inheritance
  VariableNode "$Q"
  ConceptNode "Person"

ExecutionOutputLink
  GroundedSchemaNode "deduction-formula"
  ListLink
    InheritanceLink
      VariableNode "$P"
      VariableNode "$X"
    ExecutionOutputLink
      GroundedSchemaNode "sim-to-inh-formula"
      ListLink
        InheritanceLink
          VariableNode "$P"
          VariableNode "$Q"
        SimilarityLink
          VariableNode "$P"
          VariableNode "$Q"
    InheritanceLink
      VariableNode "$Q"
      VariableNode "$X"

If P and Q are people, then one good way to evaluate if P has a certain property, is to find some similar person Q and assess whether they have that property... Which can be graphically represented by the following inference tree

Similarity P Q
-sim-to-inh-fo-
Inheritance P Q    Inheritance Q X
--------deduction-formula---------
         Inheritance P X

patterns in statistically learned NLP parses (from partially learned grammars

(Linas)

patterns in MOSES models imported to Atomspace

for a gene expression data set, genes over expressed or under expressed together in case samples could indicate a functional pathway or other ontological category differentiating cases from controls.

 (AndLink )
   (VariableNode $var_1)
   (VariableNode $var_2)
 (AndLink )
   (NotLink )
     (VariableNode $var_1)
   (NotLink )
     (VariableNode $var_2)

the third possibility is one over expressed gene and one under expressed gene, possibly indicating mutually exclusive pathways/categories

 (AndLink )
   (VariableNode $var_1)
   (NotLink )
     (VariableNode $var_2)


patterns spanning said MOSES models, and biological background knowledge

gene ontology (GO) background knowledge corresponding to the respective patterns above

 (AndLink )
   (MemberLink )
     (VariableNode $var_1)
     (ConceptNode "GO:xxx")
   (MemberLink )
     (VariableNode $var_2)
     (ConceptNode "GO:xxx")
 (AndLink )
   (MemberLink )
     (VariableNode $var_1)
     (ConceptNode "GO:yyy")
   (MemberLink )
     (VariableNode $var_2)
     (ConceptNode "GO:yyy")
 (AndLink )
   (MemberLink )
     (VariableNode $var_1)
     (ConceptNode "GO:xxx")
   (MemberLink )
     (VariableNode $var_2)
     (ConceptNode "GO:yyy")