Lab 5: The Exact Match Problem – Why Semantic Search Fails¶

Level: Intermediate | Duration: 2 hours

Objective¶

Understand the fundamental limitation of semantic embeddings and why they fail for exact matches.

The Problem¶

User searches for: "Order #1766"
Semantic closest match: "Order #1767"
Why? Both embeddings are nearly identical because the numbers look alike to the model!

What You'll Learn¶

• Create datasets that expose the problem
• Visualize why embeddings fail on numbers
• Understand semantic vs exact matching
• Recognize when to use keyword search
• Introduce solution: hybrid search
• Analyze failure cases

Real-World Scenarios Where This Happens¶

• Product SKUs: "SKU-001" vs "SKU-002"
• Order IDs: Different IDs but similar values
• Medical codes: ICD codes, diagnosis codes
• Reference numbers: Invoice numbers, tracking IDs