Fulltext search

The FulltextMatch and FulltextScore functions are intended for fulltext search in YDB via a fulltext index.

FulltextMatch

FulltextMatch(text, query) filters rows by whether the text matches a fulltext query.

SELECT id, body
FROM articles VIEW ft_idx
WHERE FulltextMatch(body, "machine learning");

Only the first two arguments can be positional. Additional parameters must be passed as named arguments.

Supports named arguments:

• Mode (String): query mode:
  • Keywords (default) — the query is split into individual terms; how they are combined is determined by DefaultOperator
  • Query — extended syntax with logical operators: required terms via +, excluded terms via -, exact phrases in double quotes
  • Wildcard — wildcard search: % matches any substring, _ matches a single character (similar to LIKE); requires an n-gram index
• DefaultOperator (String): term combination operator in Keywords mode:
  • And (default) — all query terms must be present in the text
  • Or — matching at least one term is sufficient; use MinimumShouldMatch to set a minimum threshold
• MinimumShouldMatch (String): minimum number of matched terms when DefaultOperator = "Or" — specified as an absolute number (for example, "3") or a percentage of query terms (for example, "50%")

Mode and DefaultOperator values are case-insensitive.

Wildcardexample

If the index is created with n-gram filtering, you can use % and _ patterns similar to LIKE.

SELECT id, title
FROM articles VIEW ft_idx
WHERE FulltextMatch(body, "mach% learn%", "Wildcard" AS Mode)
LIMIT 20;

MinimumShouldMatch example

An 8-term query where at least half of the terms must match:

SELECT id
FROM articles VIEW ft_idx
WHERE FulltextMatch(
  body,
  "machine learning neural networks deep learning large scale data recommendations search",
  "Keywords" AS Mode,
  "Or" AS DefaultOperator,
  "50%" AS MinimumShouldMatch
);

In this example, Mode is set to Keywords (treating the query as a set of terms), DefaultOperator is set to Or (allowing matching any subset of terms), and MinimumShouldMatch is set to "50%" (requiring at least 4 out of 8 terms to match). This helps filter out documents that contain only one or two words from a long query.

FulltextScore

FulltextScore(text, query) returns a relevance score based on the BM25 algorithm and can be used for ranking results.

Requires the fulltext_relevance index type.

SELECT id, FulltextScore(body, "machine learning") AS relevance
FROM articles VIEW ft_idx
WHERE FulltextScore(body, "machine learning") > 0
ORDER BY relevance DESC
LIMIT 10;

Only the first two arguments can be positional. Additional parameters must be passed as named arguments.

Supports named arguments:

• DefaultOperator (String): term combination operator — And (default, all terms must be present) or Or (at least one term must match)
• MinimumShouldMatch (String): when DefaultOperator = "Or", minimum number of matched terms — specified as an absolute number (for example, "2") or a percentage (for example, "50%")
• K1 (Double): term frequency saturation parameter in BM25 — controls how strongly repeated occurrences of a term affect the score; typical range: 1.2–2.0
• B (Double): document length normalization parameter in BM25 — 0.0 disables normalization, 1.0 fully normalizes by document length; typical value: 0.75

Example:

SELECT id,
       FulltextScore(body, "machine learning", "Or" AS DefaultOperator, "1" AS MinimumShouldMatch, 0.75 AS K1, 1.2 AS B) AS relevance
FROM articles VIEW ft_idx
WHERE FulltextScore(body, "machine learning", "Or" AS DefaultOperator, "1" AS MinimumShouldMatch, 0.75 AS K1, 1.2 AS B) > 0
ORDER BY relevance DESC;