The Pattern Matcher

The Pattern Matcher is the primary engine for querying and retrieving information from the Hyperon AtomSpace. It allows AI algorithms to find structured patterns within the massive hypergraph.

What is Pattern Matching?

In the context of the AtomSpace, a "pattern" is a partially completed graph (a query). The pattern matcher finds all possible ways to "fill in the blanks" (variables) using existing Atoms.

Example Query

Suppose we want to find all people who live in New York:

(Match (AtomSpace)
  (EvaluationLink
    (PredicateNode "lives-in")
    (ListLink (VariableNode "$person") (ConceptNode "New York"))
  )
  $person
)

Features of the Hyperon Pattern Matcher

1. Variables and Grounding

Supports variables that can match any Atom or specific types of Atoms.

2. High Performance

Designed to handle millions of Atoms across a distributed cluster. It uses sophisticated indexing and sub-graph isomorphism algorithms to find matches quickly.

3. Recursion and Composition

Patterns can be nested, and the results of one match can be used as input for the next. This is essential for complex reasoning.

4. Integration with MeTTa

The pattern matcher is directly exposed in the MeTTa language, making it the primary way to "read" from the system's memory.

Use Cases

• Fact Retrieval: "Who is the CEO of Google?"
• Rule Triggering: "Find all Atoms that match the preconditions of the 'buy-milk' rule."
• Knowledge Synthesis: "Find links between concept A and concept B that have a high confidence score."

Next: MeTTa Overview