6. When to Use Each Search Method

Overview

This is the most practical capsule in the module: a clear decision guide for choosing between keyword, semantic, or hybrid based on your use case.

---

Decision matrix

Use case	Recommended method	Justification
Searching for codes/IDs	Keyword	You need an exact match
Conceptual search	Semantic	Intent > exact words
Technical documentation	Hybrid	Exact technical terms + concepts
E-commerce	Hybrid	An exact product name + a description
Support/tickets	Hybrid	Exact IDs + similar descriptions
Academic research	Semantic	Concepts, not necessarily keywords
Legal/compliance	Keyword	Precision is critical
General chat/RAG	Semantic	The user's intent

---

Decision flow

┌─ Do you need EXACT matches?
│  └─ YES → Keyword
│  └─ NO → Next question
│
├─ Is the query conceptual/a question?
│  └─ YES → Semantic
│  └─ NO → Next question
│
├─ Are the queries mixed (some exact, others conceptual)?
│  └─ YES → Hybrid
│  └─ NO → Semantic (more coverage)
│
└─ Is the budget very limited?
   └─ YES → Keyword (cheaper)
   └─ NO → Semantic or hybrid

---

Concrete examples

Case 1: A personal blog search engine

Context: 500 articles about technology

Typical queries:

• "React tutorial"
• "how to learn Python"
• "testing best practices"

Decision: Semantic search

Justification:

• Conceptual queries (intent)
• A small dataset (cheap embeddings)
• Users expect relevant results, not exact ones

---

Case 2: A technical support system

Context: 100K historical tickets

Typical queries:

• "Ticket #12345" (exact)
• "problem with login" (conceptual)
• "card payment error" (conceptual)

Decision: Hybrid (keyword filter + semantic ranking)

Justification:

• Mixed queries (exact IDs + descriptions)
• Keyword for the IDs → it guarantees exactness
• Semantic for the descriptions → it finds similar tickets

---

Case 3: An e-commerce site with 1M products

Context: A large catalog

Typical queries:

• "iPhone 15 Pro" (an exact name)
• "phone with a good camera" (descriptive)
• "gaming laptop" (a category)

Decision: Hybrid (rank fusion)

Justification:

• Mixed queries
• Exact names → keyword
• Descriptions → semantic
• Fusion → an automatic balance

---

Case 4: A legal knowledge base

Context: Legal documents

Typical queries:

• "article 42 of law X"
• "cases about medical negligence"

Decision: Keyword for the exact ones, semantic for the conceptual ones

Justification:

• Precision is critical for specific articles
• Semantic is useful for finding related cases
• Separate by query type

---

Signs that you need to change

Sign 1: Users reformulate their queries several times

Symptom: A user searches for "car", finds nothing, searches for "auto", finds nothing, searches for "automobile"

Solution: Migrate to semantic search (it understands synonyms)

---

Sign 2: Irrelevant results in semantic

Symptom: The query "Python 3.11" returns docs about "python snakes"

Solution: Add a keyword filter ("Python" must appear)

---

Sign 3: You can't find things by an exact word

Symptom: You search for "error E404" but semantic returns general errors

Solution: Use keyword for exact codes/IDs

---

Summary

Key points:

• Keyword: Precision for exact matches
• Semantic: Coverage for conceptual queries
• Hybrid: A balance for mixed queries
• The decision: Based on the query types and the requirements

---

Next capsule: 07-capstone-exercise-5.md — 10 queries → decide the method.