Postgres テキスト検索ガイド 2026

SELECT * FROM posts
WHERE search_vector @@ to_tsquery('english', 'postgres & performance')
  AND category = 'tutorial'
  AND published_at > NOW() - INTERVAL '6 months';

-- Generated column keeps the index current automatically
ALTER TABLE posts ADD COLUMN search_vector tsvector
  GENERATED ALWAYS AS (
    setweight(to_tsvector('english', coalesce(title, '')), 'A') ||
    setweight(to_tsvector('english', coalesce(body,  '')), 'B')
  ) STORED;

CREATE INDEX posts_search_idx ON posts USING GIN (search_vector);

-- Query
SELECT title, ts_rank(search_vector, query) AS rank
FROM posts, to_tsquery('english', 'postgres & performance') query
WHERE search_vector @@ query
ORDER BY rank DESC
LIMIT 10;

CREATE EXTENSION IF NOT EXISTS pg_trgm;

CREATE INDEX users_name_trgm_idx ON users USING GIN (name gin_trgm_ops);

-- Finds "Micheal Jordan" when searching "Michael Jordan"
SELECT id, name, similarity(name, $1) AS score
FROM users
WHERE name % $1          -- % operator = similarity threshold (default 0.3)
ORDER BY score DESC
LIMIT 10;

-- Prefix matching for autocomplete. A trigram GIN index can help,
-- but a B-tree pattern index may be better for pure left-anchored prefixes.
SELECT name FROM users
WHERE name ILIKE $1 || '%'
ORDER BY name
LIMIT 10;

-- word_similarity for partial matches within longer strings
-- ("Johnson" within "Andrew Johnson III")
SELECT id, name, word_similarity($1, name) AS score
FROM users
WHERE $1 <% name
ORDER BY score DESC
LIMIT 10;

CREATE INDEX users_email_idx ON users (email);

-- Exact lookup: fast and unambiguous
SELECT id, name FROM users WHERE email = $1;

-- Trigram similarity on title catches typos; ts_rank handles body relevance
SELECT id, title,
  ts_rank(search_vector, to_tsquery('simple', $1)) AS fts_rank,
  similarity(title, $1) AS trgm_score
FROM posts
WHERE search_vector @@ to_tsquery('simple', $1)
   OR title % $1
ORDER BY (ts_rank(search_vector, to_tsquery('simple', $1)) + similarity(title, $1)) DESC
LIMIT 10;

-- Strip diacritical marks so "José" matches "Jose"
CREATE EXTENSION IF NOT EXISTS unaccent;

CREATE TEXT SEARCH CONFIGURATION public.simple_unaccent (COPY = pg_catalog.simple);

ALTER TEXT SEARCH CONFIGURATION public.simple_unaccent
  ALTER MAPPING FOR hword, hword_part, word
  WITH unaccent, simple;

ALTER TABLE posts ADD COLUMN search_vector tsvector;

CREATE TRIGGER posts_search_vector_refresh
BEFORE INSERT OR UPDATE OF title, body ON posts
FOR EACH ROW EXECUTE FUNCTION
  tsvector_update_trigger(search_vector, 'public.simple_unaccent', title, body);

ALTER TABLE users ADD COLUMN name_search text;

CREATE FUNCTION users_name_search_refresh()
RETURNS trigger LANGUAGE plpgsql AS $$
BEGIN
  NEW.name_search := unaccent(coalesce(NEW.name, ''));
  RETURN NEW;
END;
$$;

CREATE TRIGGER users_name_search_refresh
BEFORE INSERT OR UPDATE OF name ON users
FOR EACH ROW EXECUTE FUNCTION users_name_search_refresh();

CREATE INDEX users_name_search_trgm_idx
  ON users USING GIN (name_search gin_trgm_ops);

SELECT id, name
FROM users
WHERE name_search % unaccent($1)
ORDER BY similarity(name_search, unaccent($1)) DESC
LIMIT 10;

-- pg_search: BM25 full-text search with fuzzy matching
CREATE INDEX posts_bm25_idx ON posts
  USING bm25 (id, title, body)
  WITH (key_field = 'id', text_fields = '{"title": {}, "body": {}}');

-- Query with BM25 scoring + fuzzy matching (catches "javascipt")
SELECT id, title, paradedb.score(id) AS rank
FROM posts
WHERE posts @@@ paradedb.fuzzy_phrase(field => 'title', value => 'postgres performnce')
ORDER BY rank DESC
LIMIT 10;

-- Hybrid search: FTS + pgvector merged with RRF
WITH fts AS (
  SELECT id, ts_rank(search_vector, query) AS score,
         ROW_NUMBER() OVER (ORDER BY ts_rank(search_vector, query) DESC) AS rank
  FROM docs, to_tsquery('english', 'withRetry & timeout') query
  WHERE search_vector @@ query
  LIMIT 60
),
vec AS (
  SELECT id,
         ROW_NUMBER() OVER (ORDER BY embedding <=> $embedding) AS rank
  FROM docs
  ORDER BY embedding <=> $embedding
  LIMIT 60
)
SELECT COALESCE(fts.id, vec.id) AS id,
       (COALESCE(1.0 / (60 + fts.rank), 0) +
        COALESCE(1.0 / (60 + vec.rank), 0)) AS rrf_score
FROM fts FULL JOIN vec ON fts.id = vec.id
ORDER BY rrf_score DESC
LIMIT 10;