Text search
BM25 keyword retrieval. Rare tokens score higher; boosted fields outweigh body text.
Query syntax:
| Syntax | Meaning |
|---|---|
battery drain |
All tokens must match |
"overnight drain" |
Exact phrase |
battery OR charging |
Either token |
-firmware |
Exclude token |
subject:battery |
Restrict to field |
priority:>=high |
Numeric/date comparison |
What to index for text search:
- Titles, names, and subjects (index with a high boost)
- Short identifiers (use an exact-match analyser)
- Short descriptions
What not to index:
- Boilerplate, repeated headers, raw HTML
- Fields where meaning matters more than keywords (use vector instead)
Field boosts: a title match should outweigh a body match. Start at 2.0 for high-signal fields; keep body at 1.0. Avoid going above 5.0 — it buries relevant body content.