Back

21_Structured_KR

Loading views...
← PreviousNext →

Structured Knowledge Representation

Structured knowledge representation schemes group related information together to simplify search and inference. Unlike formal logic, which focuses on individual statements, these schemes focus on objects and their relationships.

1. Associative Networks (Semantic Nets)

Associative Networks represent knowledge as a graph of nodes (concepts) connected by labeled arcs (relationships).

Core Components:

  • Nodes: Represent objects, entities, or concepts.
  • Arcs: Represent relationships between nodes (e.g., is-a, has-part, instance-of).

Visual Example:

      [ Bird ] --- (can) ---> [ Fly ]
         |
       (is-a)
         |
      [ Sparrow ] --- (eats) ---> [ Seeds ]
         |
      (instance-of)
         |
      [ "Jack" ]

(Jack inherits "Fly" from the Bird node via the inheritance chain)

Key Concepts:

  • Inheritance: Unless specified otherwise, nodes inherit properties from their ancestor nodes (e.g., if "Bird" can "Fly", then "Sparrow" inherits "Fly").
  • Default Inheritance: Inheriting characteristics unless explicitly overridden.
  • Property Lists: A common implementation where properties are linked to a single object atom.

2. Frame Structures

Frames are data structures used to represent stereotypical situations or objects. They are similar to "classes" or "records" in programming.

Frame Components:

  • Slots: Fields representing specific attributes (e.g., Color, Age, Wife).
  • Facets: Sub-fields within slots that provide more detail:

- Value: The actual data.

- Default: Used if no value is provided.

- If-Needed: A procedure triggered to calculate a value.

  • Demons (Procedural Attachments): Procedures activated automatically when a slot is accessed or modified (if-added, if-removed, if-needed).

Example Frame: Ford Car

(Ford
  (AKO (VALUE Car))
  (Color (VALUE Silver))
  (Model (VALUE 4-door))
  (Gas-Mileage (DEFAULT 25))
  (Range (IF-NEEDED Calculate_Range))
)

3. Scripts

Scripts represent sequences of commonly occurring events in stereotypical situations, such as eating in a restaurant or shopping.

Components of a Script:

  1. Track: The specific variation of the situation (e.g., "Fast Food" vs. "Fine Dining").
  2. Roles: The actors involved (e.g., Customer, Waiter, Cashier).
  3. Props: Physical objects used (e.g., Menu, Food, Money, Table).
  4. Entry Conditions: Facts that must be true for the script to start.
  5. Scenes: Chronological sequences of events (e.g., Entering, Ordering, Eating, Paying).
  6. Results: Facts that are true after the script is completed.

4. Conceptual Dependencies (CD)

Conceptual Dependency Theory provides a limited set of primitive actions to represent any natural language statement unambiguously.

CD Primitives (ACTs):

  • ATRANS: Transfer of an abstract entity (e.g., giving money).
  • PTRANS: Physical transfer of an object (e.g., moving a chair).
  • MTRANS: Transfer of mental information (e.g., telling a story).
  • INGEST: Taking something into the body (e.g., eating).
  • PROPEL: Application of physical force (e.g., pushing).
  • MOVE: Moving a body part.
  • SPEAK: Emitting a sound.
  • ATTEND: Focusing attention on an object.

Summary Comparison

Scheme Focus Best For
Semantic Nets Relationships Hierarchical knowledge and taxonomies
Frames Object properties Stereotypical objects and entities
Scripts Sequences Stereotypical event sequences and storytelling
CD Theory Meaning Primitives Natural language understanding and semantics

5. Production Systems (Rule-Based Systems)

A Production System consists of a set of rules (productions) and a control system that decides which rules to apply.

Components:

  1. Rule Base (Knowledge Base): A set of rules in the form IF (condition) THEN (action).
  2. Working Memory: A global database of facts representing the current state of the world.
  3. Inference Engine: The control mechanism that matches rules against working memory and selects which one to "fire".

Inference Strategies:

  • Forward Chaining (Data-Driven): Starts with known facts and applies rules to derive new facts until the goal is reached.
  • Backward Chaining (Goal-Driven): Starts with the goal and works backward to see if there are facts in working memory that support it.

Conflict Resolution:

When multiple rules match the working memory, a strategy is needed to pick one:

  • Specificity: Pick the more specific rule.
  • Recency: Pick the rule that matches the most recently added facts.
  • Priority: Use pre-assigned rule weights.
← PreviousNext →