Semantic nets with example

A semantic net (semantic network) is a graph-based knowledge representation scheme. Nodes represent objects, concepts or values, and labelled arcs (links) represent relationships between them. It was introduced by Quillian and is described in Rich & Knight.

Main features

• Nodes: objects (Ram), classes (Person, Mammal), property values (Red).
• Arcs: is-a (class–subclass), instance-of (member–class), has-part, color, team, and so on.
• Property inheritance: a node inherits the attributes of its super-classes through is-a / instance links.
• Knowledge is organised around objects, so related facts can be found by following links.

Example

Facts: Tom is a cat. A cat is a mammal. A mammal is an animal. A mammal has fur. Tom's colour is black. A cat eats fish. An animal can breathe.

 [Animal] --can--> [Breathe]
    ^
    | is-a
 [Mammal] --has--> [Fur]
    ^
    | is-a
  [Cat] --eats--> [Fish]
    ^
    | instance-of
  [Tom] --colour--> [Black]

Inference using the net

• Does Tom have fur? Tom → instance-of Cat → is-a Mammal → has Fur. Yes (inherited).
• Can Tom breathe? Follow is-a up to Animal. Yes.
• What does Tom eat? Inherited from Cat: fish.

Equivalent logic

instance(Tom, Cat), isa(Cat, Mammal), isa(Mammal, Animal), ∀x instance(x,Cat) → eats(x,Fish).

Advantages

• Simple, visual and easy to understand.
• Inheritance saves storage and gives fast retrieval.
• Natural for hierarchical (taxonomic) knowledge.

Limitations

• No standard semantics for links; meaning depends on the designer.
• Hard to represent negation, disjunction, quantifiers ("all", "some").
• Exceptions (a penguin is a bird that cannot fly) need special handling.
• The net becomes large and cluttered for big domains.