This lesson's interactive features are locked, please to keep using them
Keyword vs. Semantic Search
We've built two search systems:
- Keyword Search
- Semantic Search
... but which is better? Well, it depends on the type of query:
| Query Type | Example Query | Keyword Search | Semantic Search | Winner |
|---|---|---|---|---|
| Exact title | "The Revenant" | ✓ Perfect match | ✗ May rank others higher | Keyword |
| Year | "2015" | ✓ Finds all 2015 movies | ✗ Misses without "2015" in text | Keyword |
| Concept | "family friendly" | ✗ Needs exact words | ✓ Finds Paddington, Brother Bear | Semantic |
| Theme | "survival movies" | ✗ Needs "survival" word | ✓ Understands concept | Semantic |
Keyword search results for "bear attack" might be:
- Backcountry (has "bear" + "attack")
- (no other results – too specific!)
Semantic search results for "bear attack" might be:
- The Revenant (understands "mauled by grizzly")
- Backcountry (bear attack)
- Grizzly Man (dangerous bear interactions)
Each System Has Blind Spots
Keyword search can't handle:
- Synonyms: "film" vs. "movie"
- Concepts: "uplifting" vs. "feel-good"
- Context: "bank" (river vs. financial)
Semantic search can't handle:
- Exact strings: Model numbers, IDs
- Rare terms: New slang, names
- Precision: "Toy Story 3" vs. "Toy Story"