Full-Text Search

pg_ripple provides full-text search on RDF literal objects using PostgreSQL's built-in GIN tsvector indexes. This enables fast free-text queries on string-valued predicates without scanning the entire dictionary.

fts_index

pg_ripple.fts_index(predicate TEXT) RETURNS BIGINT

Creates a GIN tsvector index on the _pg_ripple.dictionary table for string-literal objects. Returns the dictionary ID of the predicate.

The predicate argument accepts both raw IRI strings and N-Triples notation with angle brackets.

-- Create FTS index for the dc:description predicate
SELECT pg_ripple.fts_index('http://purl.org/dc/elements/1.1/description');

-- N-Triples notation also accepted
SELECT pg_ripple.fts_index('<http://purl.org/dc/elements/1.1/description>');

The index is created as IF NOT EXISTS, so calling fts_index multiple times for the same predicate is safe.

Index scope: The GIN index covers all plain string literals (kind = 2) in the dictionary. The fts_search function restricts results by predicate via a VP table JOIN.

Example: index all abstract predicates

-- Load some data
SELECT pg_ripple.load_ntriples('
    <https://example.org/paper1> <https://schema.org/abstract> "A study of RDF triplestores" .
    <https://example.org/paper2> <https://schema.org/abstract> "PostgreSQL extensions for graph data" .
');

-- Index the abstract predicate
SELECT pg_ripple.fts_index('<https://schema.org/abstract>');

-- Now search
SELECT * FROM pg_ripple.fts_search('RDF | triplestore', '<https://schema.org/abstract>');

fts_search

pg_ripple.fts_search(query TEXT, predicate TEXT)
    RETURNS TABLE(s TEXT, p TEXT, o TEXT)

Executes a full-text search against literal objects of the specified predicate. Returns matching triples as N-Triples–formatted strings.

• query — a PostgreSQL tsquery expression (see below)
• predicate — the predicate IRI, with or without angle brackets

SELECT s, o
FROM pg_ripple.fts_search('semantic & query', '<https://schema.org/abstract>');

tsquery syntax

PostgreSQL tsquery uses & (AND), | (OR), ! (NOT), and <-> (phrase proximity):

All terms are automatically stemmed (English stemmer by default) — searching for "querying" also matches "query" and "queries".

Searching without a prior fts_index call

fts_search works even without a prior fts_index call by performing a sequential scan of the dictionary joined to the VP table. For large stores, call fts_index first for each predicate you search frequently.

Return columns

Language configuration

The default text configuration is 'english', which applies English stemming and stop-word removal. If your data is in another language you can create the index manually:

-- Example: French language index on a custom predicate
CREATE INDEX my_fr_fts ON _pg_ripple.dictionary
    USING GIN (to_tsvector('french', value))
    WHERE kind = 2;

The built-in fts_index and fts_search always use 'english'. Multi-language support is planned for a future release.