Reranking¶
The Principle of Combining FTS and Vector Search¶
Reranking is the process of running two different searches, each good at different things, and then combining their strengths to produce better results than either one could alone.
In WordPress/MySQL terms:¶
- FTS (MATCH AGAINST) is great at keyword matching
- Vector search is great at semantic meaning
- Reranking blends them into one unified relevance score
This is the foundation of modern hybrid search.
Why Combine Them?¶
FTS and vector search solve different problems:
- FTS knows exact words
- Vector search knows what the words mean
Reranking lets you use both signals at once.
The Core Idea (Simple Diagram)¶
User Query
|
+------------------------------+
| |
FTS Search Vector Search
(keyword match) (semantic similarity)
| |
+--------------+---------------+
|
Combine Signals
|
Reranked Results
The system doesn’t replace one with the other — it layers them.
How Developers Should Think About It¶
FTS answers:¶
“Does this document contain the words the user typed?”
Vector search answers:¶
“Does this document mean what the user meant?”
Reranking answers:¶
“How do we blend both signals to produce the most relevant list?”
Why This Matters in WordPress/MySQL¶
1. FTS alone is brittle¶
Misspellings, synonyms, and natural language phrasing break it.
2. Vector search alone is too fuzzy¶
It may return semantically related content that doesn’t match the user’s intent.
3. Reranking gives you the best of both worlds¶
- FTS ensures precision
- Vectors ensure understanding
- Reranking ensures balance
This is exactly how modern search engines behave.
A WordPress‑Centric View¶
FTS strengths in WP¶
- Fast in MySQL
- Great for titles, slugs, product names
- Good for exact intent (“blue shoes”, “ACF tutorial”)
Vector strengths in WP¶
- Understands meaning (“how to speed up my site” ≈ “WordPress performance tips”)
- Handles long queries
- Handles conversational queries
Reranking strengths in WP¶
- Perfect for blog search, product search, documentation search
- Works beautifully with hybrid BM25 + embeddings
- Gives users “Google‑like” relevance without replacing MySQL
Summary Table¶
- FTS - Keyword precision - Ensures exact matches and strong lexical signals -
- Vector Search - Semantic understanding - Captures meaning, synonyms, intent -
- Reranking - Blends both signals - Produces the most relevant final ranking