FTS Spell Correction

MDDB includes a SymSpell-based spell checker that improves full-text search recall by correcting typos before querying the FTS index.

Enable

Spell correction is disabled by default. Enable it via environment variable:

MDDB_SPELL=true mddbd

Or with Docker:

environment: MDDB_SPELL: "true"

When enabled, MDDB builds a frequency dictionary from all indexed documents at startup. The index loads in the background — queries return 503 until ready.

Endpoints

Spell Suggestions

POST /v1/spell-suggest

Returns per-token suggestions and a corrected version of the input text.

{ "collection": "articles", "text": "quikc brwon fox", "lang": "en", "maxSuggestions": 3
}

Response:

{ "originalText": "quikc brwon fox", "suggestedText": "quick brown fox", "tokenSuggestions": [ { "token": "quikc", "suggestions": ["quick"] }, { "token": "brwon", "suggestions": ["brown"] } ]
}

Apply Corrections

POST /v1/spell-cleanup

Applies the best correction for each token and returns cleaned text.

{ "collection": "articles", "text": "quikc brwon fox", "lang": "en"
}

Response:

{ "original": "quikc brwon fox", "cleaned": "quick brown fox", "correctionsApplied": 2
}

Custom Dictionary

List — GET /v1/spell-dictionary?lang=en&collection=articles

Add — PUT /v1/spell-dictionary

{ "collection": "articles", "lang": "en", "words": ["MDDB", "mddbd", "SymSpell"], "frequency": 500
}

Higher frequency gives words stronger priority. Default is 100.

Remove — DELETE /v1/spell-dictionary

{ "collection": "articles", "lang": "en", "words": ["mddbd"]
}

Integration with FTS

CLEANED=$(curl -s -X POST http://localhost:11023/v1/spell-cleanup \ -d '{"text":"documnt retriveal","lang":"en"}' | jq -r .cleaned) curl -s -X POST http://localhost:11023/v1/search \ -d "{\"collection\":\"articles\",\"query\":\"$CLEANED\",\"lang\":\"en\"}"