Linked open government data: from five stars to EU law

Linked data is a way of publishing data on the web so that every thing it describes, such as a school, a company or a bus stop, has its own web address, and records point to related records, including those held by other organisations. For government it promises data that can be joined across agencies without a custom integration each time, which makes it one practical route to semantic interoperability. In practice it has thrived in some areas, while simpler formats such as spreadsheets and application programming interfaces (APIs) carried most open data.

The four rules of linked data

Tim Berners-Lee set out the principles in a Design Issues note on the website of the World Wide Web Consortium (W3C), dated 27 July 2006 and labelled as his personal view¹. His four rules are short enough to quote almost in full:

1. Use URIs (Uniform Resource Identifiers) as names for things.
2. Use HTTP URIs, so that people can look up those names.
3. When someone looks up a URI, provide useful information, using the standards: RDF (the Resource Description Framework) and SPARQL, its query language.
4. Include links to other URIs, so that people can discover more things.

The point of the rules is that an identifier does two jobs at once. It names something unambiguously, and anyone who looks it up gets a description that leads on to related things. A register of schools that follows them lets a planning team, an inspectorate and a transport authority all refer to the same school by the same address, instead of matching names and postcodes by hand.

The five-star scheme for open data

In 2010 Berners-Lee added a star rating to the same note to encourage people, "especially government data owners", towards good linked data¹. Each level builds on the one before, and each makes the data easier to reuse:

• one star: on the web in any format, under an open licence;
• two stars: machine-readable structured data, such as a spreadsheet rather than a scanned table;
• three stars: a non-proprietary format, such as CSV (comma-separated values) instead of Excel;
• four stars: W3C open standards (RDF and SPARQL) used to identify things, so that people can point at them;
• five stars: links to other people's data to provide context.

The first star is the one that makes data open. The note gives Creative Commons CC-BY and the UK's Open Government Licence as examples of open licences, and points out that linked data does not have to be open: plenty of it is used inside organisations¹.

RDF: data as subject, predicate and object

RDF expresses information as triples. Each triple has a subject, a predicate and an object, and a set of triples forms a graph². Predicates are always IRIs (Internationalized Resource Identifiers, a generalisation of URIs that allows a wider range of characters), subjects are IRIs or unnamed "blank nodes", and objects can also be literal values such as text, numbers and dates². The same graph can be written in several syntaxes, including Turtle (the Terse RDF Triple Language), a compact text format².

Persistent URIs for public bodies

Linked data is only as reliable as its identifiers, because other people build them into their code. W3C's Data on the Web Best Practices asks publishers to identify each dataset by "a carefully chosen, persistent URI"³. Its worked example puts transport data on a dedicated data subdomain of a city's domain and leaves the transport agency's name out of the address, because agencies get merged and renamed, while the task of running public transport continues³. The same logic applies to a council or ministry reorganisation: an address built around a function and a domain that will outlast the current structure keeps working when the organisation chart changes.

The same guidance suggests redirection services such as Permanent Identifiers for the Web or purl.org where a publisher cannot manage its own address space³. When data is removed, the URI should still answer, either with an HTTP 410 (Gone) response or with a redirect to information about where the archived copy is held³.

Open licences and EU open data law

Directive (EU) 2019/1024 on open data and the re-use of public sector information was adopted on 20 June 2019 as a recast of Directive 2003/98/EC, and Member States had to bring the laws needed to comply with it into force by 17 July 2021⁴. It asks public sector bodies to make documents available, where possible and appropriate, in formats that are "open, machine-readable, accessible, findable and re-usable, together with their metadata"⁴. Where licences are used, Member States must make standard licences available in digital form⁴.

The directive also created high-value datasets in six categories: geospatial, earth observation and environment, meteorological, statistics, companies and company ownership, and mobility⁴. These must be free of charge, with limited exceptions, machine-readable, provided through APIs and, where relevant, offered as a bulk download⁴.

The specific datasets are listed in Commission Implementing Regulation (EU) 2023/138, adopted on 21 December 2022 and published in the Official Journal on 20 January 2023⁵. It applies from 16 months after its entry into force⁵, which data.europa.eu gives as 9 June 2024⁶. The data must be offered under Creative Commons CC0 or CC BY 4.0, or an equivalent or less restrictive open licence, and labelled as high-value in its metadata⁵. Public bodies must also publish terms of use and documentation for their APIs and name a point of contact, and Member States had to report to the Commission two years after entry into force, including persistent links to the APIs for each dataset⁵.

Where linked data took off, and where simpler formats won

Neither the directive nor the regulation requires RDF. They ask for machine-readable formats, APIs and bulk downloads⁴⁵, and three-star CSV already counts as open data under Berners-Lee's own scheme¹. For tables of statistics or fast-changing operational data, a CSV file or a JSON (JavaScript Object Notation) API is usually quicker to produce and easier for developers to use.

Linked data has lasted best where shared identifiers matter more than volume: catalogue records exchanged under DCAT-AP, and the code lists and thesauri that many systems point to, covered in the guide to core vocabularies. W3C guidance takes a middle line, advising a level of formality that fits the data and its most likely uses³. Opinions still differ on whether the fourth and fifth stars repay their cost, but the two ideas underneath them, stable identifiers and links to authoritative reference data, help whatever format a publisher chooses.