ARQ Extensions: Analytics
A package for analytics that adds aggregate function builders for analyzing SPARQL result sets. These allow e.g. to obtain the set of used prefixes (for each variable or overall), used datatypes and corresponding frequency counts.
PrefixAccumulator
Depends on PatriciaTrie from org.apache.commons:commons-collections4.
Accumulate prefixes up to a capped size. Particularly useful for query optimization over virtual knowledge graphs: The prefix metadata can be used for source selection in query federation engines.
- Adding a prefix that is shorter than an existing one will “supersede” (remove) the longer one: So if “ab” exists, “a” is added then only “a” remains.
- When accumulating an item that exceeds the threshold the set of prefixes having the longest common prefix (lcp) are removed and the lcp is added.
:warning: At present this operation performs in O(n) whereas only the children of the deepest lca of the patricia trie’s leafs would have to be collapsed. TODO: Investigate whether this can be done with the PatriciaTrie API.
Example
The following example uses the core implementation of accumulator directly:
List<String> items = Arrays.asList("dbr:Leipzig", "dbr:London", "dbr:City");
PrefixAccumulator acc = new PrefixAccumulator(3);
items.stream().forEach(acc::accumulate);
System.out.println(acc.getValue());
// [dbr:City, dbr:Leipzig, dbr:London]
acc.accumulate("lgd:foo");
System.out.println(acc.getValue());
// [dbr:City, dbr:L, lgd:foo]
It is possible to wrap the accumulator as a Java8 Collector suitable for parallel procsessing. Note, that in the case of parallel processing the result will be correct but may be non-deterministic as the output of the combine step depends on the set of items the accumulators have seen w.r.t. the given threshold.
Set<String> result = Stream.of("dbr:Leipzig", "dbr:London", "dbr:City", "lgd:foo")
.collect(Aggregators.createCollector(() -> new PrefixAccumulator(3)));
System.out.println(result);
// [dbr:City, dbr:L, lgd:foo]
Several static convenince methods to work with Jena objects are provided, such as
public static Map<Var, Set<String>> analyzePrefixes(ResultSet rs, int targetSize) { }