We present ShERML, a tool for facilitating the data exchange from relational data to RDF. The main feature of the tool is a graphical user interface for designing a relational-to-RDF mapping by drawing arrows between schematic representations of the relational schema on the one hand, and a SHACL or ShEx schema on the other hand. The mapping thus constructed can be used directly by the tool for materializing the RDF graph, or be exported as an R2RML mapping.
RDF and Semantic Web data often originate from pre-existing relational
databases. The problem to export relational data to RDF is known as relational to
RDF data exchange, and can be solved either by writing ad-hoc export scripts,
or more recently by using dedicated declarative languages such as the W3C
standard R2RML. The latter allows to de ne mappings that use logical tables
(typically SQL queries) and how each row of such table is used to construct a
number of RDF triples. There exist a number of mapping editors to facilitate
the creation of R2RML mappings [
ShERML is inspired by the well known relational-to-relational data exchange
tool Clio[
Copyright c 2019 for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
The architecture of ShERML is presented in Fig. 2. The core element of ShERML is the Graphical Mapping Language (GML), which allows to de ne mappings from relational databases to RDF while specifying how the results of SQL queries should be organized to t the shapes of a shapes schema. ShERML provides a Graphical Mapping Language Editor that allows to de ne GML mappings while allowing a high degree of customization of the de ned mapping. A GML mapping can be fed into the Materializer that executes it by running SQL queries on the relational database and constructing the desired RDF graph. Additionally, the GML script can be translated to a corresponding R2RML script with the help of the Converter. The online video demo, source code and installation instructions are available on this repository https://github.com/josemachino/ShERML.
We demonstrate the principal functionality of ShERML with the following use case. We shall export the contents of a relational database with products and their suppliers as seen on the left side in Fig. 1. The database consists of three tables: Product with the information on products, Supplier with the information on suppliers, and ProdSupp with information on what supplier provides which product. The data is to be exported to RDF whose structure is de ned with the schema present on the right side in Fig. 1. In essence, it de nes two types of RDF nodes: TProduct for products and TSupplier for suppliers. The type TProduct requires a node to have a single name, a single price, and an optional origin information. Additionally, a product node must have one or more suppliers. A supplier node must have a name, a phone number, and an arbitrary number of email addresses. During the demonstration, we shall construct the desired GML mapping, in Fig. 1, with the following steps:
Loading schema. Before de ning the mapping, both schemas need to be loaded from their source les. Each table and each shape are represented with a single box, each containing the list of attributes/property names. Each element of the list has a dedicated circular anchor, for creating connections later on. Anchors use two colors: orange anchors identify elements that reference other elements i.e., foreign keys in the relational schema and shape references in shapes schema; green anchors identify remaining elements that store values.
De ning mapping. The mapping is de ned with the convenience of drawing arrows from the elements of relational schema to the elements of the shape schema. There are three kinds of arrows: IRI arrow connects relation to the corresponding shape (it does not use anchors). It de nes how rows from a table are to be mapped to IRI nodes satisfying a given shape constraint. For instance, table Product is mapped to the shape TProduct and for every row in Product a corresponding IRI is created in the output RDF using a speci ed IRI constructor TPr2iri applied to the key attribute PID (more details on IRI constructors below). Property arrow connects relation attributes to literal properties of the shapes; it can only connect two green anchors. It de nes how table rows are mapped to triples. A property arrow is dependent on an IRI arrow that de nes the IRI nodes for which the property values are generated. For instance, the arrow from Product :Name to TProduct:name maps every row in Product to a triple whose subject is TPr2iri(PID ), the predicate is name, and the object is the value of the Name attribute in the row.
Reference arrow connects a foreign key of a table to a reference property in the shape schema; it can only connect two orange anchors. Similarly to a property arrow, a reference arrow is dependent on an IRI arrow and yields RDF triples, but this time the object is also an IRI rather than a literal. For instance, the arrow from ProdSupp:SID to TProduct:suppliedBy generates the triples that connect every product with each of its suppliers.
Both property and reference arrows are dependent on an IRI arrow that speci es the subject of the generated triples. The IRI arrow does not need to originate from the same table, and those arrows are annotated with the join expressions needed to reach the relevant IRI arrow cf., the expression ProdSupp on Product on the arrow from ProdSupp:SID to TProduct:suppliedBy.
Both IRI and reference arrows are annotated with IRI constructors that specify how the IRI nodes are to be obtained. By default when loading a shape schema, the system constructs default IRI constructors for every shape in a manner consistent with how IRIs are constructed in R2RML. For instance, for the TProduct shape, whose full IRI is http:==inria:fr=TProduct, the default IRI constructor TPr2iri basically appends the given value to this IRI e.g., TPr2iri(P1) returns http:==inria:fr=TProduct=P1.
Re ning GML mapping. The mapping can then be closely inspected and rened further. For this purpose ShERML o ers a visual inspector that reorganizes the GML by grouping the arrows by the target shapes. It allows adding lter conditions and, in our demonstration we shall add a condition to the TProduct IRI arrow to export only products whose stock is lower than 1000.
RDF materialization and R2RML export. Once the GML mapping is de ned, ShERML can export the contents of the relational database to RDF in a chosen format (Turtle, N-triples, or RDF/JSON). Finally, the system allows us to convert the GML mapping to an R2RML mapping that can then be used by third-party applications.
During the demonstration we shall also present static and runtime consistency checking and property value transformations. 4
We have presented a demo of ShERML, a versatile system for de ning mappings from relational databases to RDF with target shape schema using an intuitive and powerful graphical mapping language. As next step we plan to use the tool for expressing existing relational to RDF data exchange scenarios currently written as python programs. This will allow to prioritise the future extensions of ShERML such as extending its mapping language or its GUI. Acknowledgments This work was partially supported by a grant from CPER Nord-Pas de Calais/FEDER DATA Advanced data science and technologies 2015-2020 and by the ANR project DataCert ANR-15-CE39-0009.