On this page:

SPARQL: graph patterns

A single triple pattern answers very little. Real questions need several patterns that constrain each other, and the shape they form is called a graph pattern. This chapter covers how patterns combine, how to write them without repeating yourself, and how to follow a chain of relationships you cannot count in advance.

It assumes SPARQL: the basics. All examples run against the public Pokémon dataset introduced there.

Patterns join on shared variables

A graph pattern holds zero or more triple patterns. Where two of them use the same variable, the engine keeps only the results that satisfy both. That shared variable is the join.

prefix foaf:  <http://xmlns.com/foaf/0.1/>
prefix vocab: <https://triplydb.com/academy/pokemon/vocab/>

select ?pokemon ?image ?cry
where {
  ?pokemon foaf:depiction ?image.
  ?pokemon vocab:cry ?cry.
}
limit 25

Because both patterns use ?pokemon, only Pokémon that have both a picture and a sound come back. Adding a pattern therefore narrows the result — it never widens it.

Writing the same subject once

Repeating ?pokemon on every line gets tiring. Two punctuation marks remove the repetition:

Symbol Meaning
. End of a triple pattern
; Same subject, next predicate
, Same subject and predicate, next object

The query above, written properly:

prefix foaf:  <http://xmlns.com/foaf/0.1/>
prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#>
prefix vocab: <https://triplydb.com/academy/pokemon/vocab/>

select ?pokemon ?image ?cry ?name1 ?name2
where {
  ?pokemon
    foaf:depiction ?image;
    rdfs:label ?name1, ?name2;
    vocab:cry ?cry.
}
limit 25

This is the same query as four separate patterns. The abbreviation is syntax only, and the layout — one predicate per line, indented under its subject — is worth adopting as a habit.

Filtering what comes back

filter discards results that do not meet a condition. It applies to bindings that the pattern has already produced.

prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#>
prefix vocab: <https://triplydb.com/academy/pokemon/vocab/>

select ?pokemon ?nameJa ?nameEn
where {
  ?pokemon
    vocab:name ?nameJa, ?nameEn.
  filter(lang(?nameJa) = 'ja-ja' && lang(?nameEn) = 'en-us')
}
limit 25

Two things are happening. vocab:name ?nameJa, ?nameEn matches the name property twice, so each result pairs two names for one Pokémon. The filter then keeps only the pairs where the first is Japanese and the second is English. Without it you would get every combination, including a name paired with itself.

lang() returns the language tag of a literal. Other common tests are numeric comparison (?weight > 100), inequality (?a != ?b), and regex() for text.

Making a pattern optional

Everything so far narrows. optional does not: it adds bindings when they exist and leaves them empty when they do not.

prefix foaf:  <http://xmlns.com/foaf/0.1/>
prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#>

select ?pokemon ?name ?image
where {
  ?pokemon rdfs:label ?name.
  optional { ?pokemon foaf:depiction ?image. }
}
limit 25

Every Pokémon with a name is returned. Those without a picture come back with an empty ?image cell rather than disappearing.

This matters more in linked data than in a relational database. A table forces every row to have every column; a graph does not, so absence is normal rather than exceptional.

Accepting either of two shapes

union returns results that match either pattern.

prefix rdfs: <http://www.w3.org/2000/01/rdf-schema#>
prefix skos: <http://www.w3.org/2004/02/skos/core#>

select ?resource ?label
where {
  { ?resource rdfs:label ?label. }
  union
  { ?resource skos:prefLabel ?label. }
}
limit 25

Use it when one idea is expressed two ways — which happens constantly in data that has been merged from several sources.

Supplying your own values

values injects a fixed list of bindings into a query, so you can ask about a specific set of things without writing a filter for each.

prefix vocab: <https://triplydb.com/academy/pokemon/vocab/>
prefix id:    <https://triplydb.com/academy/pokemon/id/pokemon/>

select ?pokemon ?weight
where {
  values ?pokemon { id:pikachu id:mew id:flareon }
  ?pokemon vocab:weight ?weight.
}

Property paths: following a chain

Sometimes you do not know how many steps away the answer is. A category might sit two levels up the hierarchy for one concept and five for another. Property paths express the route instead of the number of steps.

Path Meaning
P/Q Follow P, then follow Q
P\|Q Follow P or Q
P+ Follow P one or more times
P* Follow P zero or more times
P? Follow P zero or one time
^P Follow P in the opposite direction

P/Q alone removes a lot of clutter. Instead of introducing a variable for an intermediate node you will never use, write the two steps as one:

prefix rdfs:  <http://www.w3.org/2000/01/rdf-schema#>
prefix vocab: <https://triplydb.com/academy/pokemon/vocab/>

select ?pokemon ?typeName
where {
  ?pokemon vocab:type/rdfs:label ?typeName.
}
limit 25

The + and * forms are what make hierarchies queryable at all. skos:broader+ walks a taxonomy all the way to its root regardless of depth.

Property paths and OWL

The path operators line up with the OWL property characteristics you may have declared in your ontology:

If a property is Navigate it with
owl:TransitiveProperty P+
owl:ReflexiveProperty P*
owl:SymmetricProperty P\|(^P)

This is worth knowing even when your triple store does no reasoning: a property path gives you the traversal without needing inferred triples to exist.

See OWL and ontologies for what these characteristics mean.

Skipping a node you do not need

Square brackets stand for a resource you must pass through but never refer to again. Instead of inventing ?somethingIDoNotCareAbout, write:

prefix foaf:  <http://xmlns.com/foaf/0.1/>
prefix vocab: <https://triplydb.com/academy/pokemon/vocab/>

select ?image ?name
where {
  [ foaf:depiction ?image;
    vocab:name ?name ].
}
limit 25

Compact, and it makes clear which parts of the pattern carry meaning.

Where this appears in Triply products

  • The Editor navigates SKOS hierarchies, which are property paths underneath. See Edit SKOS data.

Where to next