Sparklis is a SPARQL query builder that can connect to any endpoint, and that interacts with users in natural language only. Users are guided in the building of their queries so that they do not have to know the schema, and so that empty results are almost completely avoided. This demo paper presents a number of recent extensions to Sparklis. Most notably, it now supports analytical queries, Wikidata statement quali ers, and the display of results on a map or as a slideshow.
Sparklis1 [
Sparklis is based on theoretical work that started in 2010 [
The purpose of this demonstration paper is to shortly present the main
features that have been added to Sparklis since the last publication in 2016 [
In 2016, Sparklis covered basic graph patterns (including cycles), UNION, OPTIONAL, NOT EXISTS, basic lters, basic aggregations, ORDER BY, SELECT 1 Web application and screencasts at http://www.irisa.fr/LIS/ferre/sparklis/ 2 http://data.persee.fr/explorer/sparklis/ 3 https://www.loterre.fr/sparklis/ and ASK queries. The newly covered constructs are those related to expressions (FILTER, BIND, GROUP BY, aggregators) and named graphs (FROM, GRAPH). Apart from CONSTRUCT and DESCRIBE queries and updates, which are not yet covered, only a few features are still missing: transitive closures in property paths (partly covered by hierarchies of terms, see next section), SERVICE for federated queries, explicit LIMIT and OFFSET, and separator speci cation in GROUP CONCAT. The only serious limit in expressivity is that graph patterns must be connected at all steps, for the sake of tractability of query evaluation. Expressions. SPARQL expressions, which are found in the FILTER and BIND constructs, and in the SELECT and GROUP BY clauses, bring a considerable increase in expressivity. They enable to derive values that are not explicitly present in data (e.g., people age from their birth date), and to answer analytical queries based on aggregations. Examples of analytical queries over geographical data are (permalinks to are provided as short URLs): 1. How many countries are there in Europe? (simple aggregation) http://bit.
ly/2OMxbJw 2. Give me the average population and the average area of countries, for each continent. (grouping and multiple aggregations) http://bit.ly/2OMwKPm 3. What is the average GDP per capita, for each continent? (aggregation over a BIND) http://bit.ly/2OJsErj 4. Which continents have an average agricultural GDP smaller than 10% of their average service GDP? (FILTER over two aggregations) http://bit. ly/2OLAQaM 5. Give me the distribution of the number of islands in archipelagos. (aggregation over an aggregation) http://bit.ly/2NV71QF For all those questions, an equivalent Sparklis query can be built over the MONDIAL dataset4. See the Learn/Examples page to try them live. Named Graphs. Named graphs are found in the GRAPH construct, and in FROM clauses. They allow for controling the source of triples, and are required on some endpoints that put triples in di erent named graphs. Sparklis o ers two ways to specify named graphs. The rst way is to add FROM clauses in SPARQL queries by setting two con guration options: the default graph URI for retrieving factual triples, and the schema graph URI (if di erent from the default graph) for retrieving schema triples. The second way is to produce GRAPH-constructs in SPARQL queries for a ner control. Those constructs are introduced by the new query construct according to ..., which can be applied to any subquery to declare that this subquery should be matched on a named graph. The possible named graphs are shown in the results, and the ellipsis in the construct can be re ned to select the desired named graphs. For example, the query: give me a city that according to a census whose year > 2000 has a population > 1000000 returns \the cities with over 1M inhabitants, according to a named graph that is a census posterior to 2000". 4 http://dbis.informatik.uni-goettingen.de/Mondial
Sometimes RDF datasets contain complex structures that, although covered by Sparklis 2016, generate ugly verbalizations, and are di cult to explore. We have added new constructs that abstract over those complex structures, so that they become more intuitive to query and explore.
Rei ed Relationships. Because RDF is limited to binary relationships
(properties), a common solution to represent n-ary relationships or to add metadata
to binary relations is to reify those relationships into statements or events.
Unfortunately, various representations are used without consensus (e.g., standard
rei cation, custom rei cation, singleton properties, named graphs, Wikidata
predicates [
Hierarchies of Terms. Some properties code for the child-to-parent
relationship of a hierarchy. Their range is equal to their domain, and in SPARQL queries
it is usually their transitive closure that is used. General-purpose and well-known
examples are rdfs:subClassOf or skos:broader. Domain-speci c hierarchies
exist for geographical locations, historical periods, anatomical parts, etc. We
have added the construct (hierarchy) in ... that uses the last selected
property as a hierarchical relationship. The e ects are to apply transitive closure
to the property in the generated SPARQL query, and to display the values
of that property as a hierarchical tree. When a value in the tree is selected,
it is equivalent to select all values under the selected value in the tree. It is
possible to select several values and combine them by and/or. For example,
the verbose and approximate query give me every place that is the part
of the part of California or Arizona (assuming two levels between places
and states) can now be expressed as give me every place (hierarchy) in
California or Arizona.
User experience is important and we have also made improvements on this side.
{ Responsive design. The UI has been given a responsive design with
Bootstrap. It becomes therefore easier to use Sparklis on smaller screens.
{ Map view. When entities in the query have standard properties for
latitude/longitude, they are detected and a special property has geolocation
is suggested. If selected, this property triggers the display of the entities in
the query results on a Google map (ex., http://bit.ly/2NT2Vs9).
{ Slideshow view. When the results contain images, those are collected
and displayed as a slideshow. To provide context for the images inside the
slideshow, the result row that contains the image is displayed under the
image. For example, in DBpedia, a slideshow of scientist depictions can be
obtained, along with biographic information http://bit.ly/2NRSwwT.
{ Full-text search. In order to provide a better experience when using the
dynamic ltering features, support has been added for the full-text search
capabilities of a few RDF stores (Jena/fuseki and Virtuoso so far). It assumes that
full-text indices have been set up on the endpoint. For example, cities with
name similar to \petersbourg" are retrieved by http://bit.ly/2OH8sqg.
{ YASGUI editor. For advanced users who want to see and edit the
generated SPARQL query, the well-known YASGUI editor [
Spanish, and Dutch. Similar languages would be easy to add.