We present Shape Designer, a graphical tool for building SHACL or ShEx constraints for an existing RDF dataset. Shape Designer allows to automatically extract complex constraints that are satis ed by the data. Its integrated shape editor and validator allow expert users to combine and modify these constraints in order to build an arbitrarily complex ShEx or SHACL schema.
A Shape Designer project is associated with an RDF graph G and a ShEx or SHACL schema under construction S. The language to use for the schema, ShEx or SHACL, is xed at project creation. The GUI of Shape Designer is shown in Fig. 1. It includes the following panels: (1) a list of shape patterns; (2) a list of node selection queries; (3) an editing area containing the schema under construction S; (4) a validation view (not shown in Fig. 1).
The main functionality of Shape Designer is the automatic construction of a (ShEx or SHACL) shape constraint. For a given node selection query Q and a shape pattern P , it creates a shape constraint constr (Q; P ) that has a structure as indicated by pattern P and that is satis ed by all the nodes of G in the result of query Q. The constraint thus obtained can be added to the schema under construction, or simply used to gather information about the data. 3
We demonstrate how to construct a recursive ShEx schema S that has one constraint for every rdfs:Class used in G and accounts for the references between types. We use a dataset with uniform structure from DBTunes4. Create the project. At project creation we choose the shape language, here ShEx, and indicate to the tool where to nd the graph: in a local le, a local RDF4J database, or through a SPARQL endpoint. A new project comes with a set of prede ned useful patterns. For instance, pattern P1 = f rdf:type [ ] ; ~ g is prede ned and indicates that the constraint to be constructed should contain the list of possible values ([ ]) for property rdf:type, and a datatype or 4downloaded from http://dbtune.org/classical/ on June 16, 2019.
Shape Designer for ShEx and SHACL Constraints demo:Composition { a [ mo:Composition ] ; mo:composer [ composer:~ ]; mo:produced_work IRI * ; bio:date xsd:integer ? ; bio:place xsd:string ? } :Shape_QComp_P2 { mo:composer {
a [ type:Composer ] } mo:produced_work {
a [ mo:MusicalWork ] } * } ; node kind value constraint ( ) for all other properties (~). We also ask the tool to construct queries that select all nodes of some class, for all the classes that appear as objects of rdf:type in the graph. For instance, query QComp = SELECT ?x WHERE f?x a mo:Compositiong is generated by the tool. Note that the pre xes de ned in the graph can be used in queries and in patterns. Automatic construction of shapes. Now, given P1 and QComp, the tool inspects the graph and constructs constr (QComp; P1) shown on the left hand side of Fig. 2. Remark that the cardinality of each triple constraint was automatically inferred from the data, among four possible cardinality constraints that are exactly one, optional (?), at least one (+), and any number (*). We can choose to add constr (QComp; P1) to the schema S under construction. By repeating the construction of constr (QC ; P1) for all classes C, a novice user can build a schema that models the data without need of mastering the shape constraint language. For instance, we create shapes demo:Composer and demo:MusicalWork for the nodes with classes type:Composer and mo:MusicalWork, respectively. Validation. At any time the user can validate graph G against the schema under construction S. The validation result is shown in a list and allows to explore the neighbourhood of nodes to help understand validation errors.
Shape patterns to explore the structure of the data. Automatic shape construction can be used to gather information about the data. Consider pattern P2 = f mo:~ f rdf:type [ ] g g. It indicates that we are interested only in properties with namespace mo:, and we want to retrieve the lists of rdf:types of their values. The shape constr (QComp; P2) is presented on the right-hand side of Fig. 2 and it shows that the values of property mo:composer have class type:Composer, and the values of property mo:produced work have class mo:MusicalWork. Editing the schema. Automatically inferred schemas use only subsets of the ShEx and SHACL languages. For instance, they never use shape references. We can create complex schemas by editing the schema under construction. For instance, using the information gathered from constr (QComp; P2) we can introduce shape references in demo:Composition in Fig. 2 as follows.
mo:composer @demo:Composer ; mo:produced_work @demo:MusicalWork *
Wikidata is (partially) crowd-sourced, thus highly heterogeneous. This is an essential di erence compared to the rst use case in which nodes with same rdf:type have very similar properties. Shape Designer can be used to understand the structure of Wikidata entries and assert the quality of such entries for a particular application in mind. We assume the reader familiar with the basic concepts of Wikidata (direct properties, property statements, quali ers).
Shape Designer o ers special support for Wikidata in the form of Wikidata project type that comes with prede ned pre xes and patterns. Shape construction and validation are performed by querying the Wikidata SPARQL endpoint, therefore we put a limit on the number of query results. Validating the whole Wikidata set or even all entities of given type is computationally prohibitive. In this context the most useful feature of Shape Designer are patterns, as they allow to focus on a particular kind of information in Wikidata. For instance, a simple pattern for direct properties allows to get an idea on which properties can be expected for entities of given type. More complex patterns can account for the presence of absence of provenance information. 5
We have presented Shape Designer, a versatile tool for constructing shape
schemas or exploring RDF datasets designed to be usable by novices and experts.
Shape Designer uses ideas similar to several prototype tools for extracting ShEx
or SHACL constraints from RDF graphs or from Wikidata [
Shape Designer is being actively developed, in particular by adding more functionality for exploring Wikidata. In the near future we plan to extend shape patterns and the schema construction algorithm so that even more complex schemas could be built automatically, including recursive shapes. Acknowledgments This work was partially funded by a grant from CPER Nord-Pas de Calais/FEDER DATA Advanced data science and technologies 2015-2020, by the ANR project DataCert ANR-15-CE39-0009, by the Spanish Ministry of Economy and competitiveness (Society Challenges: TIN2017-88877-R), and by the \Severo Ochoa" research program (BP17-88).