KR models; semantic nets

Knowledge representation (KR) is the way facts about the world are encoded so that a program can store them and reason with them. A good scheme should have representational adequacy, inferential adequacy, inferential efficiency and acquisitional efficiency (Rich & Knight).

Knowledge representation models

1. Logical representation: propositional logic and predicate logic (FOPL). Precise semantics and sound inference (e.g. resolution), but can be inefficient. Example: ∀x man(x) → mortal(x).
2. Production rules: IF condition THEN action. Used in expert systems; modular and easy to explain. Example: IF fever AND rash THEN measles.
3. Semantic networks: graph of nodes and labelled arcs with inheritance.
4. Frames: slot-and-filler structures with defaults and procedures.
5. Conceptual dependency (scripts): language-independent meaning using primitive acts; scripts describe event sequences (restaurant script).
6. Procedural representation: knowledge embedded as procedures or programs (e.g. Prolog clauses, LISP functions).
7. Probabilistic / fuzzy models: Bayesian networks, fuzzy sets for uncertain knowledge.

Semantic nets

A semantic net is a labelled graph: nodes are objects or concepts, arcs are relations (is-a, instance, has, colour). Properties are inherited through is-a links.

Example: "Rani is a student. A student is a person. A person has a head. Rani studies at Pulchowk."

 [Person] --has-part--> [Head]
    ^ is-a
 [Student]
    ^ instance
 [Rani] --studies-at--> [Pulchowk]

Query "Does Rani have a head?" is answered by moving Rani → Student → Person → Head.

Advantages: easy to understand, supports inheritance, compact. Limitations: no standard meaning for links, weak for negation, quantifiers and exceptions, and large nets are hard to manage.