Structured Knowledge Representation: Semantic Networks — Beginner-Friendly Notes

1. What It Is

Semantic network = graph of knowledge: nodes = concepts/objects (Bird, Tweety, Feathers), edges = relations (IsA, HasPart, InstanceOf). Reasoning is graph traversal / inheritance.

Logic is formula-centric; semantic nets are graph/visual-centric, closer to how humans picture categories.

2. Basic Building Blocks

• IsA / Subclass: $Canary \xrightarrow{IsA} Bird$ (all canaries are birds)
• Instance: $Tweety \xrightarrow{InstanceOf} Canary$ (Tweety is a specific canary)
• Property: $Bird \xrightarrow{HasProperty} Feathers$, $Bird \xrightarrow{Can} Fly$

Inheritance: Tweety inherits properties of Canary, Bird via IsA links — $Tweety$ can fly because $Canary \rightarrow Bird \rightarrow CanFly$.

3. Tiny Diagram (text)

Bird ──Has──► Feathers
 ▲ Can
 │
Canary ──Instance── Tweety

Query "Does Tweety fly?" → traverse $Tweety \rightarrow Canary \rightarrow Bird \rightarrow CanFly$ → yes.

4. Reasoning & Inference

• Inheritance: follow IsA links upward to find properties.
• Exception handling: add cancellation — e.g., $Penguin \xrightarrow{IsA} Bird$ but $Penguin \xrightarrow{NotCan} Fly$ overrides inheritance (needs default reasoning).
• Spreading activation: start at nodes and spread along edges to find related concepts (used in language).

5. Advantages vs Logic

Semantic Nets	Logic
Visual, intuitive, fast inheritance	Formal, precise, general quantifiers
Good for taxonomies / hierarchies	Good for arbitrary rules
Weak semantics (what does edge exactly mean?)	Clear semantics

Historical note: used in early AI (Quillian's TLC), still lives inside knowledge graphs (Google Knowledge Graph, WordNet).

6. Limitations (exam point)

• No clear semantics: edge meaning ambiguous compared to FOL ($\forall x\ (Canary(x) \Rightarrow Bird(x))$ is precise).
• Cannot easily represent: negation, disjunction, quantification (e.g., "every bird..." needs logic).
• Fix: convert to FOL or use more formal description logics.

7. One-Line Summary

Semantic networks encode knowledge as nodes and labelled edges with inheritance along IsA/InstanceOf links, offering intuitive taxonomic reasoning but with weaker formal semantics than logic.

8. Quick Self-Check

1. How does Tweety inherit "flies" from Bird?
2. Why is "Penguin is a bird but doesn't fly" a problem for simple nets?