SPARQL Aggregation

Properties#

• Every selected variable is grouped or aggregated. A variable that is not in GROUP BY (like ?paper) has many values per group. It must either be left out of the SELECT or appear inside an aggregate.
• That is why the lecture’s query groups by ?venue ?name and not just ?venue: ?name is selected. If every venue has exactly one label, both give the same groups.
• The aggregate is a new variable and can be used afterwards, e.g. in ORDER BY DESC(?num_papers) (Ordering).

Sub-SELECT. A complete SELECT … WHERE { … } GROUP BY … in braces can stand inside a WHERE clause. Its result table is joined with the other patterns of the outer query via their shared variables, like a Subquery in SQL’s FROM.

• Aggregate first, then join. In the lecture, the inner query counts papers per venue, and the outer query adds ?venue rdfs:label ?name afterwards. The result is the same as with the flat query, but cheaper: the label is joined once per venue, instead of once per paper before grouping.
• Inside the sub-SELECT, only the variables in its SELECT (here ?venue, ?num_papers) are visible to the outer query.

Example#

Toy data: papers $p_1, p_2$ in venue $V$, paper $p_3$ in venue $W$, each venue with one label.

SELECT ?venue ?name (COUNT(?paper) AS ?num_papers) WHERE {
  ?paper dblp:publishedInStream ?venue .
  ?venue rdfs:label ?name .
}
GROUP BY ?venue ?name
ORDER BY DESC(?num_papers)

The WHERE clause gives 3 rows $(p_1, V, \text{name}_V)$, $(p_2, V, \text{name}_V)$, $(p_3, W, \text{name}_W)$. Grouping gives $(V, \text{name}_V, 2)$, $(W, \text{name}_W, 1)$.

Sub-SELECT variant (Slide 22):

SELECT ?venue ?name ?num_papers WHERE {
  {
    SELECT ?venue (COUNT(?paper) AS ?num_papers) WHERE {
      ?paper dblp:publishedInStream ?venue .
    } GROUP BY ?venue
  }
  ?venue rdfs:label ?name .
}
ORDER BY DESC(?num_papers)

The inner query gives $(V, 2)$, $(W, 1)$. Only these 2 rows are joined with the labels, instead of the 3 paper rows.

Sources#

• Slides 01, p. 21 (GROUP BY, COUNT, rule for non-grouped variables), p. 22 (sub-SELECT)
• ADIS_VL01 > GROUP BY and Aggregation, ADIS_VL01 > Sub-SELECT