Q1. Scenario: You are building a medical diagnosis system for rare diseases. How would you represent medical knowledge (facts and rules)? What are the advantages of using an ontology?
Use a knowledge base with frames or description logic. Represent symptoms, diseases, and relationships. Ontology allows hierarchical classification (e.g., disease subtypes), inheritance, and reasoning consistency.
Q2. Scenario: A chatbot needs to understand that "Paris is the capital of France" and "France is a country in Europe". Represent these facts using predicate logic. Then answer: "What is the capital of a country in Europe?"
Facts: capital(France, Paris); country(France); country_in(France, Europe). Query: ∃c ∃x (capital(c,x) ∧ country_in(c, Europe)) → x = Paris. Use inference by backward chaining.
Q3. Scenario: Compare semantic networks and frames. Give an example where frames are more expressive.
Semantic networks are graphs of nodes (concepts) and edges (relations). Frames add structure: slots with default values, procedural attachments, and inheritance. Example: frame "Car" has slots "make", "model", and "drive()" method.
Q4. Scenario: You have a rule-based expert system for loan approval. Explain forward chaining and backward chaining. Which is more suitable for goal-driven diagnosis?
Forward chaining: start with facts (credit score, income) and apply rules to derive new facts. Backward chaining: start from goal (approve loan) and work backwards to find supporting evidence. Backward chaining suits diagnosis (goal-driven).
Q5. Scenario: A knowledge graph for a product recommendation system. What types of relationships would you store? How can you use it to recommend "customers who bought X also bought Y"?
Relationships: product_category, bought_together, similar_to, belongs_to. For "also bought", traverse the graph: find all products co-purchased with X via transaction edges, then aggregate frequency.