W3C

Use Cases and Requirements for Mapping Relational Databases to RDF

W3C Working Draft 2 June 2010

This version:
http://www.w3.org/TR/2010/WD-rdb2rdf-ucr-20100602/
Latest version:
http://www.w3.org/TR/rdb2rdf-ucr/
Editors:
Eric Prud'hommeaux, W3C <eric@w3.org>
Michael Hausenblas, DERI, NUI Galway <michael.hausenblas@deri.org>
Authors:
Sören Auer, Universität Leipzig <auer@informatik.uni-leipzig.de>
Lee Feigenbaum, Cambridge Semantics <lee@thefigtrees.net>
Daniel Miranker, University of Texas at Austin <miranker@cs.utexas.edu>
Angela Fogarolli, University of Trento <afogarol@disi.unitn.it>
Juan Sequeda, University of Texas at Austin <jsequeda@cs.utexas.edu>

Abstract

The need to share data with collaborators motivates custodians and users of relational databases (RDB) to expose relational data on the Web of Data. This document examines a set of use cases from science and industry, taking relational data and exposing it in patterns conforming to shared RDF schemata. These use cases expose a set of functional requirements for exposing relational data as RDF in the RDB2RDF Mapping Language (R2RML).

Status of this Document

This section describes the status of this document at the time of its publication. Other documents may supersede this document. A list of current W3C publications and the latest revision of this technical report can be found in the W3C technical reports index at http://www.w3.org/TR/.

This is the First Public Working Draft of the "Use Cases and Requirements for Mapping Relational Databases to RDF" for review by W3C members and other interested parties.

Comments on this document should be sent to public-rdb2rdf-comments@w3.org, a mailing list with a public archive.

Publication as a Working Draft does not imply endorsement by the W3C Membership. This is a draft document and may be updated, replaced or obsoleted by other documents at any time. It is inappropriate to cite this document as other than work in progress.

The W3C RDB2RDF Working Group is the W3C working group responsible for this document.

This document was produced by a group operating under the 5 February 2004 W3C Patent Policy. W3C maintains a public list of any patent disclosures made in connection with the deliverables of the group; that page also includes instructions for disclosing a patent. An individual who has actual knowledge of a patent which the individual believes contains Essential Claim(s) must disclose the information in accordance with section 6 of the W3C Patent Policy.

Scope

Per the RDB2RDF charter, the scope of the RDB2RDF Mapping Language (R2RML) is limited to read-only access to the relational database. This means that data can be extracted from the relational database, but not updated.

Table of Contents

1 Introduction
   1.1 Why Mapping RDBs to RDF?
   1.2 Why a standard RDB2RDF method?
2 Use Cases
   2.1 UC1 - Patient Recruitment
         2.1.1 Queries Over the RDF Graph
         2.1.2 Derived Requirements
   2.2 UC2 - Web applications (Wordpress)
         2.2.1 Queries Over the RDF Graph
         2.2.2 Derived Requirements
   2.3 UC3 - Integrating Enterprise Relational databases for tax control
         2.3.1 Querying
         2.3.2 Derived Requirements
   2.4 UC4 - rCAD: RNA Comparative Analysis Database
         2.4.1 Expose Relational Data as RDF when there is no existing Domain Ontology to map the relational schema to
         2.4.2 Expose Mapping a Relational Database to an OWL DL ontology
         2.4.3 Derived Requirements
3 Requirements
   3.1 Core Requirements
         3.1.1 DIRECT - Direct Mapping
         3.1.2 TRANSFORM - Transformative Mapping
         3.1.3 GUIDGEN - Generation of Globally Unique Identifiers
         3.1.4 SQLGEN - Query Translation
         3.1.5 ETL - Extract-Transform-Load
         3.1.6 DATATYPES - Datatypes
         3.1.7 VSDATATYPES - Database Vendor Specific Datatypes
         3.1.8 RENAMECOL - Ability to Rename SQL Column Names
         3.1.9 APPLYFUNCTION - Apply a Function before Mapping
         3.1.10 MANYTOMANY - Exposing many-to-many join tables as simple triples
         3.1.11 CLASSESFROMATTRIBVALUES - Creating Classes based on Attribute Values
   3.2 Optional Requirements
         3.2.1 NAMEDGRAPH - Named Graphs
         3.2.2 NSDECL - Namespace declaration
         3.2.3 METADATA - Static Metadata
         3.2.4 PROVENANCE - Provenance
         3.2.5 UPDATES - Update Logs
4 Acknowledgments
5 References

Appendices

A Glossary (Non-Normative)
B RDB2RDF Mapping Approaches (Non-Normative)
B.1 Direct Mapping
B.2 Direct Mapping Plus Ontology Mapping
B.3 Database to Ontology Mapping


1 Introduction

The majority of dynamic Web content is backed by relational databases (RDB), and so are many enterprise systems [DynaWebSites]. On the other hand, in order to expose structured data on the Web, Resource Description Framework (RDF) [RDF] is used. This document reviews use cases and requirements for a relational database to RDF mapping (RDB2RDF) with the following structure:

  1. The remainder of this section motivates why mapping RDBs to RDF is necessary and needed and highlights the importance of a standard.
  2. In the next section RDB2RDF use cases are reviewed.
  3. The last section discusses requirements regarding a RDB2RDF mapping language, driven by an analysis of the aforementioned use cases.

1.1 Why Mapping RDBs to RDF?

The Web of Data is constantly growing due to its compelling potential of facilitating data integration and retrieval. At the same time however, RDB systems host a vast amount of structured data in relational tables augmented with integrity constraints. In order to make this huge amount of relational data available for the Web of Data, a connection must be established between RDBs and a format suitable for the Web of Data.

The advantages of creating an RDF view of relational data are inherited from the Web of Data and can be summarized based on the tasks they facilitate:

  • Integration: data in different RDBs can be integrated using RDF semantics and mechanisms; in this sense, the Web of Data can be imagined as one big database. Moreover, information in the database can be integrated with information that comes from other data sources.
  • Retrieval: once data are published in the Web of Data (as opposed to relational databases), queries can span different data sources and more powerful retrieval methods can be built.

RDF data in the Web should be defined and linked in a way that makes it accessible for humans and machines [LinkedData]. The Web of data is a scalable environment with explicit semantics where not only humans can navigate information, but also machines are able to find connections and use them to navigate through the information space. In order to realise this global information space, we need to:

  • follow a common model to describe, connect and access resources;
  • name resources in an unambiguous way

The most common way to publish resources in the Web of Data follows the RDF model [RDF] and uses Uniform Resource Identifiers ([URI]) for resource identification, thereby facilitating the creation of a comprehensive and flexible resource description.

In the following, we will use RDB2RDF to denote any technique that takes as an input a RDB (schema and data) and produces one or more RDF graphs, as depicted in the following figure:

RDB2RDF principle

The consumer of the RDF Graph (virtual or materialized) essentially can access the RDF data in different ways:

  1. Query access, which means the agent issues a SPARQL query against an endpoint exposed by the system and processes the results (typically the result is a SPARQL result set in XML or JSON);
  2. Entity-level access, which means the agent performs an HTTP GET on a URI exposed by the system and processes the result (typically the result is an RDF graph);
  3. Dump access, which means the agent performs an HTTP GET on dump of the entire RDF graph, for example in Extract, Transform, and Load (ETL) processes.

1.2 Why a standard RDB2RDF method?

With the advent of the Web of Data, researcher and practitioners have compared the RDB model and the RDF model [RDB-RDF] and surveyed different approaches to map them [RDBMSMapping]. As noted in the survey on existing RDB2RDF mapping approaches [RDB2RDFSurvey], a couple of proposals on how to tackle the RDB2RDF mapping issues are known.

Use of a standard for mapping language for RDB to RDF may allow use of a single mapping specification in the context of mirroring of schema and (possibly some or all of the) data in various databases, possibly from different vendors (e.g., Oracle database, MySQL, etc.) and located at various sites. Similarly structured data (that is, data stored using same schema) is useful in many different organizations often located in different parts of the world. These organizations may employ databases from different vendors due to one or more of many possible factors (such as, licensing cost, resource constraints, availability of useful tools and applications and of appropriate database administrators, etc.). Presence of a standard RDB2RDF mapping language allows creation and use of a single mapping specification against each of the hosting databases to present a single (virtual or materialized) RDF view of the relational data hosted in those databases and this RDF view can then be queried by applications using SPARQL query or protocol.

Another reason for a standard is to allow easy migration between different systems. Just as a single web-page in HTML can be viewed by two different Web browsers from different vendors, a single RDB2RDF mapping standard should allow a user from one database to expose their data as RDF, and then, when they export their data to another database, allow the newly imported data to be queried as RDF without changing the mapping file. For example, imagine that a database administrator is exposing weather data as Linked Data to be consumed by other applications. At first, this weather data is stored in a light-weight database (such as MySQL). However, as more and more weather data is collected, and more and more users access the Web data, the light-weight database may have difficulty scaling. Therefore, the database administrator migrates their database to a more heavy-weight database (such as Oracle Database 11g). Of course, the database administrator does not want to re-create the ability to view the data as RDF using a vendor-specific mapping file, but instead wants to seamlessly migrate the view of their data as RDF.

A standardized mapping between relational data and RDF allows the database administrator to migrate the view of their data as RDF across databases, allowing the vendors to compete on functionality and features rather than forcing database administrators to rewrite their entire relational data to RDF mapping when they want to migrate their data from one database to another.

Another motivation for a standard is that for certain classes of systems (such as CMS) a 'default' mapping could be defined which can be deployed no matter what underlying RDB is used. As these systems, such as Drupal or Wordpress, can be run on top of different underlying relational databases. A standardized way of mapping between relational data and RDF hence allows the underlying database to be changed without disturbing the content management system.

Further, having a standard mapping would simplify programming applications that access multiple database sources.

2 Use Cases

With integration of relational data from one RDB with various kinds of data (such as relational, spreadsheets, CSV, unstructured text, etc.), the use cases presented in the following fall into one or more of the following categories:

  1. I want to integrate my RDB with another structured source (RDB, XLS, CSV, etc.), so I'll convert my RDB to RDF and assume my other structured source can also be in RDF. See UC1, UC3.
  2. I want to integrate my RDB with existing RDF on the web (Linked Data), so I'll convert my RDB to RDF and then I'm able to link and integrate. See UC2, UC4.
  3. I want to integrate my RDB with unstructured data (HTML, PDF, etc.), so I'll convert my RDB to RDF and assume my other unstructured source can also be in RDF.
  4. I want my RDB data to be available for SPARQL or other RDF-based querying, and/or for others to integrate with other data sources (structured, RDF, unstructured). See UC2, UC4.

2.1 UC1 - Patient Recruitment

The Semantic Web for Health Care and Life Sciences Interest Group has created several demonstrators using SPARQL to query clinical and biological relational databases. Included is the database structure, sample data, and a SPARQL query. Following are six tables of sample diabetic patient data extracted from the University of Texas Health Science Center. Some columns have been omitted from this use case for brevity.

Accompanying each table are two RDF views (represented in Turtle) corresponding to HL7/RIM and CDISK SDTM ontology in RDFS. While there are many motivations for providing a common interface to administer distinct databases (access to patient history, shared rules for clinical decision support, etc.), in this case, SPARQL queries (following the table description) were used to find candidates for clinical studies. For these RDF graphs, the following namespaces apply:

@prefix xsd: <http://www.w3.org/2001/XMLSchema#> .
@prefix hl7: <http://www.hl7.org/v3ballot/xml/infrastructure/vocabulary/vocabulary#>  .
@prefix stdm: <http://www.sdtm.org/vocabulary#>  .

Person
ID SexDE DateOfBirth LastEditedDTTM
1234561 2 1983-01-02 00:00:00 2007-11-13 15:49:20
1234562 3 1963-12-27 00:00:00 2008-01-30 17:08:42
1234563 2 1983-02-25 00:00:00 2007-03-10 06:01:55
<http://hospital.example/DB/Person/ID.1234561#record> a hl7:Person ;
   hl7:administrativeGenderCodePrintName Sex_DE:M ;
   hl7:livingSubjectBirthTime "1983-01-02T00:00:00Z"^^xsd:dateTime .

<http://hospital.example/DB/Person/ID.1234562#record> a hl7:Person ;
   hl7:administrativeGenderCodePrintName Sex_DE:F ;
   hl7:livingSubjectBirthTime "1963-12-27T00:00:00Z"^^xsd:dateTime .
<http://hospital.example/DB/Person/ID.1234563#record> a hl7:Person ;
   hl7:administrativeGenderCodePrintName Sex_DE:M ;
   hl7:livingSubjectBirthTime "1983-02-25T00:00:00Z"^^xsd:dateTime .
<http://hospital.example/DB/Person/ID.1234561#record> a sdtm:Patient ;
  stdm:dateTimeOfBirth "1983-01-02T00:00:00Z"^^xsd:dateTime .
<http://hospital.example/DB/Person/ID.1234562#record> a sdtm:Patient ;
  sdtm:dateTimeOfBirth "1963-12-27T00:00:00Z"^^xsd:dateTime ;
  sdtm:sex <http://hospital.example/DB/Sex_DE#F> .
<http://hospital.example/DB/Person/ID.1234563#record> a sdtm:Patient ;
  sdtm:dateTimeOfBirth "1983-02-25T00:00:00Z"^^xsd:dateTime ;
  sdtm:sex <http://hospital.example/DB/Sex_DE#F> .

Sex_DE
ID EntryCode EntryName EntryMnemonic
2 1 Male M
3 2 Female F
<http://hospital.example/DB/Zex_DE/ID.2#record>

	hl7:administrativeGenderCodePrintName "Male"@en-us ;
	Sex_DE:EntryMnemonic "M"@en-us .
<http://hospital.example/DB/Zex_DE/ID.3#record>
	hl7:administrativeGenderCodePrintName "Female"@en-us ;
	Sex_DE:EntryMnemonic "F"@en-us .
<http://hospital.example/DB/Person/ID.1234561#record>
	stdm:sex <http://hospital.example/DB/Sex_DE#M> .
<http://hospital.example/DB/Person/ID.1234562#record>
	stdm:sex <http://hospital.example/DB/Sex_DE#F> .

<http://hospital.example/DB/Person/ID.1234563#record>
	stdm:sex <http://hospital.example/DB/Sex_DE#M> .

Item_Medication
ID PatientID ItemType PerformedDTTM
99999999002 1234561 ME 2007-09-28 00:00:00
99999999003 1234562 ME 2007-09-28 00:00:00
99999999004 1234562 ME 2008-07-28 00:00:00
<http://hospital.example/DB/Person/ID.1234561#record>

	hl7:substanceAdministration <http://hospital.example/DB/Item_Medication/ID.99999999002#record> .
<http://hospital.example/DB/Item_Medication/ID.99999999002#record>
	a hl7:SubstanceAdministration ;
	hl7:effectiveTime _:t1 .
_:t1 hl7:start "2007-09-28T00:00:00"^^xsd:dateTime .
<http://hospital.example/DB/Person/ID.1234562#record>

	hl7:substanceAdministration <http://hospital.example/DB/Item_Medication/ID.99999999003#record> ;
	hl7:substanceAdministration <http://hospital.example/DB/Item_Medication/ID.99999999004#record> .

<http://hospital.example/DB/Item_Medication/ID.99999999003#record>
	a hl7:SubstanceAdministration ;
	hl7:effectiveTime _:t2 .
_:t2 hl7:start "2007-09-28T00:00:00"^^xsd:dateTime .
<http://hospital.example/DB/Item_Medication/ID.99999999004#record>
	a hl7:SubstanceAdministration ;
	hl7:effectiveTime _:t3 .
_:t3 hl7:start "2008-07-28T00:00:00"^^xsd:dateTime .

Medication
ID ItemID Dose Refill QuantityToDispense DaysToTake PrescribedByID MedDictDE
88888888002 99999999002 2 6 180 45 1004682 132139
88888888003 99999999002 2 0 180 45 1004683 132139
88888888004 99999999003 2 6 180 45 1004682 132139
88888888005 99999999004 4 6 180 45 1004682 132139
<http://hospital.example/DB/Item_Medication/ID.99999999002#record>

	hl7:consumable <http://hospital.example/DB/Medication_DE/ID.132139#record> .
_:t1 hl7:durationInDays 45 .
<http://hospital.example/DB/Item_Medication/ID.99999999003#record>

	hl7:consumable <http://hospital.example/DB/Medication_DE/ID.132139#record> .
<http://hospital.example/DB/Item_Medication/ID.99999999004#record>
	hl7:consumable <http://hospital.example/DB/Medication_DE/ID.132139#record> .
_:t2 hl7:durationInDays 45 .

<http://hospital.example/DB/Item_Medication/ID.99999999005#record>
	hl7:consumable <http://hospital.example/DB/Medication_DE/ID.132139#record> .
_:t3 hl7:durationInDays 45 .

Medication_DE
ID Entry EntryCode EntryName NDC Strength Form UnitOfMeasure DrugName DisplayName
132139 131933 98630 GlipiZIDE-Metformin HCl 2.5-250 MG Tablet 54868079500 2.5-250 TABS MG GlipiZIDE-Metformin HCl GlipiZIDE-Metformin HCl 2.5-250 MG Tablet
<http://hospital.example/DB/Medication_DE/ID.132139#record>

	hl7:displayName "GlipiZIDE-Metformin HCl 2.5-250 MG Tablet" .
NDCcodes
ingredient RxCUI labelType name NDC
6809 351273 Clinical Glipizide 2.5 MG / Metformin 500 MG Oral Tablet 54868079500
<http://hospital.example/DB/Medication_DE/ID.132139#record>
	spl:activeIngredient _:i1 .
_:i1 spl:classCode 54868079500 .

[ a sdtm:ConcomitantMedication ;
	sdtm:subject <http://hospital.example/DB/Person/ID.1234561#record> ;
	sdtm:standardizedMedicationName "GlipiZIDE-Metformin HCl 2.5-250 MG Tablet" ;
	hl7:activeIngredient [hl7:classCode 54868079500 ] ;
	sdtm:startDateTimeOfMedication "2007-09-28 00:00:00"^^xsd:dateTime ] .

[ a sdtm:ConcomitantMedication ;
	sdtm:subject <http://hospital.example/DB/Person/ID.1234562#record> ;
	sdtm:standardizedMedicationName "GlipiZIDE-Metformin HCl 2.5-250 MG Tablet" ;
	hl7:activeIngredient [ hl7:classCode 54868079500 ] ;
	sdtm:startDateTimeOfMedication "2007-09-28 00:00:00"^^xsd:dateTime ] .

[ a sdtm:ConcomitantMedication ;
	sdtm:subject <http://hospital.example/DB/Person/ID.1234562#record> ;
	sdtm:standardizedMedicationName "GlipiZIDE-Metformin HCl 2.5-250 MG Tablet" ;    
	hl7:activeIngredient [ hl7:classCode 54868079500 ] ;
	sdtm:startDateTimeOfMedication "2008-07-28 00:00:00"^^xsd:dateTime ] .

2.1.1 Queries Over the RDF Graph

The use case can be realised as a SPARQL query over the RDF graphs. The following is a SPARQL query, which extracts patients taking a particular class of medication, an anticoagulant, and not another (weight loss, here).

PREFIX sdtm: <http://www.sdtm.org/vocabulary#>
PREFIX spl: <http://www.hl7.org/v3ballot/xml/infrastructure/vocabulary/vocabulary#>

SELECT ?patient ?dob ?sex # ?takes ?indicDate ?indicEnd ?contra
 WHERE {
  ?patient a sdtm:Patient ;
          sdtm:middleName ?middleName ;
          sdtm:dateTimeOfBirth ?dob ;
          sdtm:sex ?sex .

  [   sdtm:subject ?patient ;
      sdtm:standardizedMedicationName ?takes ;

      spl:activeIngredient [ spl:classCode 6809 ] ;
          sdtm:startDateTimeOfMedication ?indicDate
  ] .
  OPTIONAL {
  [   sdtm:subject ?patient ;
      sdtm:standardizedMedicationName ?disqual ;
      spl:activeIngredient [ spl:classCode 11289 ]
          sdtm:startDateTimeOfMedication ?indicDate
  ] }
} LIMIT 30

2.1.2 Derived Requirements

This use case leads to the following requirements:

  • Map relational data to an RDF schema derived from the relational schema.
  • Create a virtual RDF view that is used to translate SPARQL queries over the RDF views to SQL queries over the relational data.
  • Mapping of relational data types to RDF/XML datatypes, for example, SQL date/time formats to XML Schema date/time formats.
  • Mapping column names to RDF property names, for example, the column DaysToTake from the Medication table is mapped to hl7:durationInDays.

2.2 UC2 - Web applications (Wordpress)

In order to make the Web of Data useful to ordinary Web users, RDF and OWL have to be deployed on the Web on a much larger scale. Web applications such as Content Management Systems, online shops or community applications (e.g. Wikis, blogs, forums) already store their data in relational databases. Providing a standardized way to map the relational data and schema behind these Web applications into RDF, RDF-Schema and OWL will facilitate novel semantic browsing and search applications.

By supporting the long tail of Web applications and thus counteracting the centralization of the Web 2.0 applications, the planned RDB2RDF standardization will help to give control over data back to end-users and thus promote a democratization of the Web.

To support this use case, the mapping language should be easily implementable for lightweight Web applications and have a shallow learning curve to foster early adoption by Web developers.

We illustrate this use case with the example of Wordpress. Wordpress is a popular blogging Web application and installed on tens of thousands of Web servers. Wordpress used a relational database (MySQL) with a relatively simple schema:

Wordpress SQL Schema

Wordpress SQL Schema:

A mapping should be able to reuse existing vocabularies.

Mapped to RDF the resulting ontology should contain the classes post, attachment, tag, category, user and comment. An example instance of the post class, for example, should look like:

@prefix sioc: <http://rdfs.org/sioc/ns#> .
@prefix dc: <http://purl.org/dc/terms/> .
@prefix dc11: <http://purl.org/dc/elements/1.1/> .

<http://blog.aksw.org/triplify/post/8> a sioc:Post .

<http://blog.aksw.org/triplify/post/8> sioc:has_creator <http://blog.aksw.org/triplify/user/5> .
<http://blog.aksw.org/triplify/post/8> dc:created       "2007-02-27T17:23:36"^^<http://www.w3.org/2001/XMLSchema#dateTime> .

<http://blog.aksw.org/triplify/post/8> dc11:title       "Submissions open: 3rd Workshop on Scripting for the Semantic Web" .

<http://blog.aksw.org/triplify/post/8> sioc:content     "The submissions web-site is now open for the 3rd Workshop on Scripting for the Semantic Web..." .
<http://blog.aksw.org/triplify/post/8> dc:modified      "2008-02-22T21:41:00"^^<http://www.w3.org/2001/XMLSchema#dateTime> .

2.2.1 Queries Over the RDF Graph

An example SPARQL query on the resulting ontology could be:

SELECT ?name, ?title
WHERE {
  ?post   rdf:type       sioc:Post .
  ?post   dc:title       ?title .
  ?post   dc:has_creator ?author
  ?author foaf:name      ?name
}

2.2.2 Derived Requirements

This use case leads to the following requirements:

  • Support Extract-Transform-Load (ETL) for the created RDF by the mapping.

2.3 UC3 - Integrating Enterprise Relational databases for tax control

Integrating relational databases and exposing them on the Web or Intranet requires the re-use of unique identifiers in order to integrate and interlink data about entities on different databases.

The re-use of unique identifiers allows:

  • Joins between entities described in different databases.
  • Join structured data (SQL) to structured data, from incompatible schema, or where data is dirty, poorly normalized, lacking proper keys/indices.

This use case is a pilot project for the Trentino region tax agency. Trentino is an autonomous region in the north of Italy. The region has a population of 1 million people and more than 200 municipalities with their own information systems. The goal of is to integrate and link tax related data about people, organizations, buildings, etc. This data come from different databases especially from the region���s many municipalities, each with their own individual data structures.

The re-use of unique identifiers will provide a lightweight method for aggregating the data. In this way we are providing a tax agent an intelligent tool for navigating through the data present in the many different databases. Using unique identifiers, a tool can aggregate data and create a profile for each tax payer. Each user profile shows different type of information, with links to other entities such as the buildings owned, payments made, location of residence etc.

The RDF generated from the two databases is materialized and joined using the generated unique identifiers.

Example

Supposed that we have two tables (Anagrafe and Urban_Cadastre) from different databases, we select some typical attributes for the two tables to explain our conversion method. Table Angrafe includes the information about two type of entities, persons and locations (a person���s residence place), and some other information:

Firstname Lastname City_Residence_Place Country_Residence_Place Other_Info
Paolo Bouquet Trento Italy xyz...

Table Urban_Cadastre contains the information about buildings and their owners:

Owner_Name Building_LocalID Building_Address Building_Type
Paolo Bouquet 123456 VIA G.LEOPARDI 3

DDL:

CREATE TABLE Anagrafe (
 Firstname varchar
 Lastname varchar
 City_Residence_Place varchar
 Country_Residence_Place varchar
 Other_Info varchar
 PRIMARY KEY (Firstname, Lastname)
)

CREATE TABLE (
 Owner_Name varchar
 Building_LocalID integer
 Building_Address varchar
 Building_Type varchar
 PRIMARY KEY (Owner_Name)
)
                

Using traditional RDB2RDF translation methods the RDF representation for the two example tables coming from two different databases is shown below:

@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix database_anagrafe: <http://www.database1.org/anagrafe/> .

database_anagrafe:entry_row1 database_anagrafe:Other_Info "xyz" ;
    database_anagrafe:Country_Residence_Place "Italy" ;
    database_anagrafe:City_Residence_Place "Trento" ;
    database_anagrafe:Last_Name "Bouquet" ;
    database_anagrafe:First_Name "Paolo" .
@prefix rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#> .
@prefix database_urbano: <http://www.database1.org/Urban_Cadastre/> .

database_urbano:entry_row1 database_urbano:Building_Type "3" ;
    database_urbano:Building_Address "VIA G.LEOPARDI" ;
    database_urbano:Building_LocalID "123456" ;
    database_urbano:Owner_Name "Paolo Bouquet" .

2.3.1 Querying

If we wanted to query these two tables, we would have to create a unique identifier, such as http://www.example.org/Paolo_Bouquet for Paolo Bouquet, to refer to entities in order to join descriptions about the same entity coming from different data sources.

2.3.2 Derived Requirements

This use case leads to the following requirements:

  • Create unique identifiers for the entities described by the data.

2.4 UC4 - rCAD: RNA Comparative Analysis Database

rCAD - RNA Comparative Analysis using SQLServer: the tremendous increase in available biological information creates opportunities to decipher the structure, function and evolution of cellular components while presenting new computational challenges for performance and scalability. To fully utilize this large increase in knowledge, it must be organized for efficient retrieval and integrated for multi-dimensional analysis. Given this, biologists are able to invent new comparative sequence analysis protocols that will yield new and different structural and functional information. Based on Microsoft SQL-server, we have designed and implemented the RNA Comparative Analysis Database - rCAD which supports comparative analysis of RNA sequence and structure, and unites, for the first time in a single environment, multiple dimensions of information necessary for alignment viewing, sequence metadata, structural annotations, structure prediction studies, structural statistics of different motifs, and phylogenetic analysis. This system provides a queryable environment that hosts efficient updates and rich analytics.

For this use-case, we will be presenting only the Sequence Alignment schema. The SQL-DDL for Microsoft SQL Server can be found here: rCAD Sequence Alignment SQL DDL

This use-case presents the issue of a schema that does not have an existing domain ontology in which it can be mapped to. The closest domain ontology is the Multiple Alignment Ontology (MAO) which can only be mapped to the Sequence Alignment part of the entire rCAD database. However, MAO is in the OBO language. Nevertheless, OBO ontologies can be translated to OWL (and back).

2.4.1 Expose Relational Data as RDF when there is no existing Domain Ontology to map the relational schema to

Rob from the RNA lab would like to expose the Sequence Alignment data from the rCAD database as RDF. However, there is no existing domain ontology in which the relational schema can be mapped to. Therefore, an ontology should be derived automatically from the relational schema. The RDF data will become instance of this automatically generated ontology.

The Alignment table from the rCAD database is the following:

CREATE TABLE [AlignmentClassic].[Alignment]  ( 
    [AlnID]             int NOT NULL,
    [SeqTypeID]         tinyint NOT NULL,
    [AlignmentName]     varchar(max) NULL,
    [ParentAlnID]       int NULL,
    [NextColumnNumber]  int NOT NULL,
    CONSTRAINT [PK_Alignment] PRIMARY KEY([AlnID])
)
ON [PRIMARY]
GO
ALTER TABLE [AlignmentClassic].[Alignment]
    ADD CONSTRAINT [FK_Alignment_SequenceType]
    FOREIGN KEY([SeqTypeID])
    REFERENCES [dbo].[SequenceType]([SeqTypeID])
    ON DELETE NO ACTION 
    ON UPDATE NO ACTION 
GO
ALTER TABLE [AlignmentClassic].[Alignment]
    ADD CONSTRAINT [FK_Alignment_Alignment]
    FOREIGN KEY([ParentAlnID])
    REFERENCES [AlignmentClassic].[Alignment]([AlnID])
    ON DELETE NO ACTION 
    ON UPDATE NO ACTION 
GO

and the desired ontology that is generated automatically from the relational schema is the following:

PREFIX rcad: <http://rcad.org/vocabulary/rcad.owl#>
rcad:Alignment rdf:type owl:Class .
rcad:AlignmentName rdf:type owl:DatatypeProperty.
rcad:AlignmentName rdfs:domain rcad:Alignment.
rcad:AlignmentName rdfs:range xsd:string.
rcad:NextColumnNumber rdf:type owl:DatatypeProperty.
rcad:NextColumnNumber rdfs:domain rcad:Alignment.
rcad:NextColumnNumber rdfs:range xsd:integer.
rcad:ParentAlnID rdf:type owl:ObjectProperty
rcad:ParentAlnID rdfs:domain rcad:Alignment
rcad:ParentAlnID rdfs:range rcad:Alignment

and the desired RDF triples, which are instances of the automatically generated ontology are the following:

PREFIX rcad: <http://rcad.org/vocabulary/rcad.owl#>

PREFIX rcad-data: <http://rcad.org/vocabulary/rcad-data.rdf#>

rcad-data:alignment1000 a rcad:Alignment;
    rcad:AlignmentName "My Alignment 1000"^^xsd:string;
    rcad:NextColumnNumber "123"^^xsd:int;
    rcad:ParentAlnID rcad-data:alignment2000

2.4.2 Expose Mapping a Relational Database to an OWL DL ontology

Rob, from the RNA lab would like to expose the Sequence Alignment data from the rCAD database as RDF. Just recently the Multiple Alignment Ontology (MAO) as been released, which is an "ontology for data retrieval and exchange in the fields of multiple DNA/RNA alignment, protein sequence and protein structure alignment." However, this ontology has been developed in OBO. Nevertheless, OBO ontologies can be translated to OWL ontologies, specifically OWL DL. Therefore, Rob will like to map his rCAD database to the MAO ontology

The Alignment and Alignment Column tables from the rCAD database is the following:

CREATE TABLE [AlignmentClassic].[Alignment]  ( 
    [AlnID]             int NOT NULL,
    [SeqTypeID]         tinyint NOT NULL,
    [AlignmentName]     varchar(max) NULL,
    [ParentAlnID]       int NULL,
    [NextColumnNumber]  int NOT NULL,
    CONSTRAINT [PK_Alignment] PRIMARY KEY([AlnID])
)
GO
ALTER TABLE [AlignmentClassic].[Alignment]
    ADD CONSTRAINT [FK_Alignment_SequenceType]
    FOREIGN KEY([SeqTypeID])
    REFERENCES [dbo].[SequenceType]([SeqTypeID])
    ON DELETE NO ACTION 
    ON UPDATE NO ACTION 
GO
ALTER TABLE [AlignmentClassic].[Alignment]
    ADD CONSTRAINT [FK_Alignment_Alignment]
    FOREIGN KEY([ParentAlnID])
    REFERENCES [AlignmentClassic].[Alignment]([AlnID])
    ON DELETE NO ACTION 
    ON UPDATE NO ACTION 
GO

---

CREATE TABLE [AlignmentClassic].[AlignmentColumn]  ( 
    [AlnID]         int NOT NULL,
    [ColumnNumber]  int NOT NULL,
    [ColumnOrdinal] int NOT NULL,
    CONSTRAINT [PK_AlignmentColumn] PRIMARY KEY([AlnID],[ColumnNumber])
)
GO
ALTER TABLE [AlignmentClassic].[AlignmentColumn]
    ADD CONSTRAINT [FK_AlignmentColumn_Alignment]
    FOREIGN KEY([AlnID])
    REFERENCES [AlignmentClassic].[Alignment]([AlnID])
    ON DELETE NO ACTION 
    ON UPDATE NO ACTION 
GO

This is just part of the Multiple Alignment Ontology in OWL DL.

2.4.3 Derived Requirements

This use case leads to the following requirements:

  • Map relational data to an RDF schema derived from the relational schema.
  • Map from the local ontology to a domain ontology.
  • Allow transformation of SPARQL queries over the RDF view into efficient SQL queries.

3 Requirements

Following is a set of proposed requirements for R2RML.

3.1 Core Requirements

The analysis of the above use cases yields a set of core requirements for the R2RML.

3.1.1 DIRECT - Direct Mapping

Relational schema and data are a potentially cyclic graph where nodes are tables or tuples and edges are either foreign/primary key relationships or table attributes. This relational graph can be directly mapped to a RDF graph where the nodes of the relational graph correspond to the subject or object and the edges of the relational graph correspond to the predicate. This directly mapped RDF graph represents exactly the information in the relational database.

The relational schema can be directly mapped to a RDFS/OWL ontology while the relational data is directly mapped to a RDF graph, which is an instance of the RDFS/OWL ontology - this ontology is considered the local ontology. This local ontology can be used when it is desired to let the database schema determine the effective ontology of the RDF view. An example of direct mapping is shown in Section 3.1.

A minimal configuration MUST provide a (virtual) RDF graph representing the attributes and relationships between tuples in the relational database.

This requirement comes from UC1 and UC2 as well as the first part of UC4.

3.1.2 TRANSFORM - Transformative Mapping

It is good Web of Data practice to re-use existing domain ontologies. Mapping between the relational graph or the local ontology with a domain ontology usually requires graph transformations [GraphTransform]. An example of this mapping is shown in Section 3.2. The local ontology considers the teacher classification (Math, Physics, etc) as literal values while in the domain ontology the teacher classifications are RDFS/OWL classes.

The R2RML language MUST express transformations of the relational graph to produce the (virtual) RDF graph.

This requirement comes from the second part of UC4.

3.1.3 GUIDGEN - Generation of Globally Unique Identifiers

RDF identifiers for objects in the conceptual model can, in some cases, be generated from a transformation of the schema and data in a tuple representing that conceptual model. For example, it may be sufficient to identify a patient in a clinical database with primary key patientID and value 1234561 as http://myclinic.example/patient/patientID.12334561#x, while but if the patient IDs are shared with another database, it will be necessary to transform one or both of these identifiers into a common form, e.g. http://allclinics.example/sharedRecords/patient.12334561#y. These use cases are labeled custom-identifier.

Given a row in a protein database with a primary key attribute "ID" and another unique attribute "uniProt":

ID uniProt name seqLength
18 68250 YYHAB 246 AA

Note:

The RDB2RDF would like to ask the world which of the following subject mappings are likely to meet the most use cases:
  <http://mydb.example/prots/ID=18> db:name "YYHAB" .       # consistent function of the primary key
  <http://mydb.example/prots18/more/path> db:name "YYHAB" . # user-defined function of the primary key
  <http://www.uniprot.org/uniprot/P68250> db:name "YYHAB" . # user-defined function of arbitrary attributes

The former uses a potentially hard-coded formula, the middle uses a user-supplied function of the primary key and the latter uses a function of a different attribute to produce a common proteomic node label.

The R2RML language MUST allow for a mechanism to generate globally unique identifiers for database entities. The generation of identifiers should be designed to support the implementation of the Linked Data principles [LinkedData]. Where possible, R2RML SHOULD encourage the reuse of public identifiers for long-lived entities such as persons, corporations and geo-locations.

This requirement comes from UC3, but also found necessary in other use cases.

3.1.4 SQLGEN - Query Translation

One use of R2RML is to materialize RDF views of data. Another is to define virtual RDF views, enabled by translating SPARQL queries over the RDF view into SQL queries over the original database.

The R2RML language MUST allow mapping specification to have sufficient information to enable transformation of SPARQL queries over the RDF view into efficient SQL queries over the relational database.

This requirement comes from the second part of UC4.

3.1.5 ETL - Extract-Transform-Load

In certain cases query access or entity-level access (see also explanation in Section 1.1 Why Mapping RDBs to RDF? ) to the exposed RDF graph is not sufficient.

The R2RML language MUST allow the mapping to have sufficient information to provide a dump of the entire RDF graph (materialzed RDF graph).

This requirement comes from UC2, but in fact all use cases require this.

3.1.6 DATATYPES - Datatypes

Relational data types MUST be treated consistently with RDF datatypes per SQL-XSD mapping ISO IWD 9075-14:2011(E) Subclause 9.5, "Mapping SQL data types to XML Schema data types", possibly including XML datatypes defined at http://standards.iso.org/iso/9075/2003/sqlxml.

This requirement comes from UC1, but in fact all use cases require this.

3.1.7 VSDATATYPES - Database Vendor Specific Datatypes

The R2RML language MUST provide an extensibility mechanism to allow for mapping database vendor-specific data types.

3.1.8 RENAMECOL - Ability to Rename SQL Column Names

When mapping from a relational schema to an RDF Schema or OWL ontology and column names are mapped to property names, it MUST be possible to rename the property names.

This requirement comes from UC1.

3.1.9 APPLYFUNCTION - Apply a Function before Mapping

In some cases, what you want to map is not the original value in the database but the result of applying a function to the value. For example, the value may be temperature in Centigrade and you may want to convert to Fahrenheit. Or from Euros to Dollars. It is easy to think of other examples. Position may be stored as a tuple (latitude, longitude and you may want to map to two separate properties. Or the address may be stored in a number of columns and you may want to map as a single string. A more complex example: a database row might contain Wiki text, which should be transformed into HTML. This can be achieved, for example, by using standard or user defined SQL functions or using XQuery/XPath Functions and Operators.

The mapping language SHOULD allow for applying a function before mapping.

3.1.10 MANYTOMANY - Exposing many-to-many join tables as simple triples

Relational databases typically use three tables to represent Many-to-Many relationships. This requirements is to allow such relationships to be mapped using direct links between the entities.

TeacherID TeacherName
1 Adams
2 Baker
3 Clark
... ...
StudentId TeacherID
1 1
1 2
2 3
3 2
... ...
StudentId StudentName
1 Davis
2 Evans
3 Frank
... ...

The community's school district maintains an RDB with basic student and personnel information, including a STUDENTS and a TEACHERS table. The relationship between the two is given in a STUDENT_TEACHER table:

The SQL DDL for these tables are the following:

CREATE TABLE student (
  studentID int PRIMARY KEY,
  studentName varchar
)

CREATE TABLE teacher (
  teacherID int PRIMARY KEY,
  teacherName varchar
)

CREATE TABLE student_teacher (
  studentID int,
  teacherID int,
  PRIMARY KEY(studentID, teacherID),
  FOREIGN KEY(studentID) REFERENCES student(studentID), 
  FOREIGN KEY(teacherID) REFERENCES teacher(teacherID)
)

SemantEducaTrix, the most recent Semantic Web company to burst into the educational software market, is mapping the school system's relational database to RDF/SPARQL. They'd like to access the relationships modelled with this join table as simple links between students and teachers:

 ex:student1 ex:studentName "Davis".
 ex:student2 ex:studentName "Evans".
 ex:student2 ex:studentName "Frank".
                
 ex:teacher1 ex:teacherName "Adams".
 ex:teacher2 ex:teacherName "Baker".
 ex:teacher3 ex:teacherName "Clark".
                
 ex:student1 ex:has_teacher ex:teacher1, ex:teacher 2 ;
 ex:student2 ex:has_teacher ex:teacher3 ;
 ex:student3 ex:has_teacher ex:teacher2 ;
 ...
                

3.1.11 CLASSESFROMATTRIBVALUES - Creating Classes based on Attribute Values

The requirement is for creating multiple classes from a single column based on the values of a related attribute.

The TEACHERS table, see above, has a Classification column:

TeacherId Classification ...
1 History
2 Physics
3 Music

SemantEducaTrix wants to instantiate these teachers as different {{{rdf:type}}}s depending on the value in the Classification column:

 ex:teacher1 a ex:HistoryTeacher .
                 ex:teacher2 a ex:PhysicsTeacher .
                 ex:teacher3 a ex:MusicTeacher .
                

3.2 Optional Requirements

3.2.1 NAMEDGRAPH - Named Graphs

The mapping language SHOULD enable the creation of multiple Named Graphs within one mapping definition.

3.2.2 NSDECL - Namespace declaration

The mapping language SHOULD enable the declaration and use of namespace prefixes.

3.2.3 METADATA - Static Metadata

The mapping language SHOULD enable the attachment of static metadata (such as licensing information) to all RDF entities or instances of a certain class. This is in particular important, when RDF is published as Linked Data on the Web, for example when one wants to state that the dataset at hand is available under a certain license, such as the Public Domain Dedication and License.

Editorial note
See also next requirement which could be considered a special case of static metadata.

3.2.4 PROVENANCE - Provenance

The mapping language SHOULD support the preservation of provenance by generating additional RDF triples according to a provenance vocabulary, such as http://purl.org/net/provenance/.

3.2.5 UPDATES - Update Logs

The mapping language SHOULD provide extension points to support the creation of update logs of relational data.

Editorial note
Not really in the scope of R2RML, hence a nice-to-have feature.

4 Acknowledgments

The editors gratefully acknowledge contributions from the members of the W3C RDB2RDF Working Group: Marcelo Arenas, Sören Auer, Samir Batla, Richard Cyganiak, Daniel Daniel Miranker, Souripriya Das, Alexander de Leon, Alexander de Leon, Orri Erling, Ahmed Ezzat, Lee Feigenbaum, Angela Fogarolli, Enrico Franconi, Howard Greenblatt, Wolfgang Halb, Harry Halpin, Nuno Lopes, Li Ma, Ashok Malhotra, Ivan Mikhailov, Juan Sequeda, Seema Sundara, Ben Szekely, Edward Thomas, and Boris Villazón-Terrazas.

5 References

DynaWebSites
A survey on dynamic Web content generation and delivery techniques, Jayashree Ravi, Zhifeng Yu, Weisong Shi, 2009. (See http://www.cs.wayne.edu/~weisong/papers/ravi09-dynamic-content.pdf.)
LinkedData
Linked Data, Tim Berners-Lee, Design Issue Note, 2006. (See http://www.w3.org/DesignIssues/LinkedData.html.)
SPARQL
SPARQL Query Language for RDF, Eric Prud'hommeaux and Andy Seaborne 2008. (See http://www.w3.org/TR/rdf-sparql-query/.)
SQL
ISO (International Organization for Standardization). ISO/IEC 9075-2:1999, Information technology --- Database languages --- SQL --- Part 2: Foundation (SQL/Foundation). [Geneva]: International Organization for Standardization, 1999. (See http://www.iso.ch/cate/d26197.html.)
RDF-RDB
RDF Access to Relational Databases (See http://www.w3.org/2003/01/21-RDF-RDB-access/.)
RDB-RDF
Relational Databases on the Semantic Web, Tim Berners-Lee, Design Issue Note, 1998. (See http://www.w3.org/DesignIssues/RDB-RDF.html.)
RDBMSMapping
SWAD-Europe Deliverable 10.2: Mapping Semantic Web Data with RDBMSes, Dave Beckett and Jan Grant, 2003. (See http://www.w3.org/2001/sw/Europe/reports/scalable_rdbms_mapping_report/#sec-mapping.)
RDB2RDFSurvey
A Survey of Current Approaches for Mapping of Relational Databases to RDF, Satya S. Sahoo, Wolfgang Halb, Sebastian Hellmann, Kingsley Idehen, Ted Thibodeau Jr, Sören Auer, Juan Sequeda, Ahmed Ezzat, 2009. (See http://www.w3.org/2005/Incubator/rdb2rdf/RDB2RDF_SurveyReport.pdf.)
GraphTransform
Wikipedia Article on Graph rewriting (See http://en.wikipedia.org/wiki/Graph_rewriting.)
RDF
Resource Description Framework (RDF): Concepts and Abstract Syntax, G. Klyne, J. J. Carroll, Editors, W3C Recommendation, 10 February 2004 (See http://www.w3.org/TR/2004/REC-rdf-concepts-20040210/.)
ReuseableIDs
Reusable Identifiers in the RDB2RDF mapping language, Michael Hausenblas and Themis Palpanas, 2009. (See http://esw.w3.org/topic/Rdb2RdfXG/ReusableIdentifier.)
URI
RFC3986 - Uniform Resource Identifier (URI): Generic Syntax (See http://tools.ietf.org/html/rfc3986.)

A Glossary (Non-Normative)

domain ontology [Definition: an ontology that has been developed by experts in the domain and accepted by a community (for example, FOAF, SIOC, Gene Ontology, etc.)]

graph [Definition: TBD]

entity [Definition: TBD]

federation [Definition: TBD]

identifier [Definition: TBD]

label [Definition: TBD]

local ontology [Definition: an ontology that has been derived from the relational schema]

named graph [Definition: a graph that is given a URI, see [SPARQL] for details.]

mapping [Definition: TBD]

vocabulary [Definition: see ontology]

B RDB2RDF Mapping Approaches (Non-Normative)

The RDB2RDF Working Group would like to call out, without specifically endorsing, three emergent approaches to mapping relational data to RDF.

B.1 Direct Mapping

Supplying a relational database (schema and data) plus a stem URI defines an RDF graph, which emulates the relational schema.

B.2 Direct Mapping Plus Ontology Mapping

The RDB2RDF may define a mapping semantics as a mapping to a direct graph, followed by the application of RDF graph transformations into an RDF graph in a final ontology.

B.3 Database to Ontology Mapping

The RDB2RDF may define a mapping semantics as a single step process from database to an RDF graph in a final ontology.