Re-user’s Guide to the Open Data Rights Statement Vocabulary

Mon Dec 23, 2013
$download_content = get_field('download_content');

This guide provides some advice on how to use metadata published in the ODRS vocabulary within your applications to help simplify the process of attributing and citing the sources of your data

This guide discusses how to process machine-readable rights statements that have been published using the Open Data Rights Statement vocabulary (ODRS).

Data published to the web should always be accompanied by machine-readable metadata that describes the dataset, its means of creation and, importantly, a statement of the rights that describes how the data can be re-used. As a re-user of Open Data you should ensure that your usage conforms to those rights. This includes making efforts to ensure that you properly attribute your sources.

The Publishers Guide to the Open Data Rights Statement Vocabulary, provides an introduction to the ODRS vocabulary and includes a number of worked examples. You should read that guide to familiarise yourself with the basic concepts behind the vocabulary.

The ODRS github project includes some additional standalone examples that can be used a further reference.

This guide provides some advice on how to use metadata published in the ODRS vocabulary within your applications to help simplify the process of attributing and citing the sources of your data.

Discovering Rights Statements

Publishers should be exposing structured metadata about their datasets using standard formats such as the Data Catalog vocabulary (DCAT) or VoID.

As described in the publisher guide there are two standard properties that can be used to reference licensing and rights information:

  • dct:license — existing practice is for this property to refer to a licence that applies to the dataset
  • dct:rights — this property can be used to refer to a odrs:RightsStatement resource that provides a fuller description of a Rights Statement, e.g. licences that refer to both content and data

In most cases these properties will be the starting point for discovering machine-readable rights metadata.

odrs:RightsStatement resource may have several properties that can be useful in your application:

  • The odrs:copyrightNotice and odrs:copyrightStatement properties respectively provide the text or a link to a copyright statement that applies to the dataset you are using. Some licences, including those from the Creative Commons, require you to preserve copyright notices provided by the data publisher. These properties allow you to include a short notice in your application or a link to a fuller copyright statement.
  • The attributionText and attributionURL properties provide the means for publishers to describe their preferred form of attribution

A Rights Statement might be published alongside a description of a dataset, e.g. as embedded metadata expressed using RDFa. However they might also be published separately, e.g. as a standalone web page or document that is referenced by several datasets.

Rights information can also be referenced directly from individual documents, e.g. using the atom:link element or in an HTML linkelement.

Web APIs might also include HTTP Link Headers, as defined by RFC 5988, that refer to rights information, e.g:

Link:; rel=”license”,; rel=””

In short, while the structure of a rights statement is defined by the ODRS vocabulary there are several different locations in which references to that metadata might be included, either along a description of the dataset or by reference. When processing the data in your application you should ensure to fetch any additional resources.

Rather than hard-coding this metadata into your application you should consider using the data to drive the creation of attribution links and citations. A publisher may need to revise the text or link target and by relying on the (cached) metadata you will be able to use the latest values in your application.

Attributing Datasets

The publisher guide includes some background discussion on attribution and citation. While closely related, there are two slightly different use cases:

  • Attribution — giving credit to the creator and/or maintainer of a dataset in order to acknowledge that their contribution to your application
  • Citation — linking to a source dataset that you have used in your application, or some analysis, or which are components of a large aggregated dataset that you might be distributing to others

Attribution is often a legal requirement that is included in Open Data licences. This means that you must attribute your sources to stay within the terms of the licence. In addition you may need to preserve any copyright notices that are provided by the publisher.

Citation is often more detailed than attribution. A clear provenance statement, e.g. a list of data sources, can help users trust your application or attempt to reproduce your analysis.

The ODRS vocabulary includes properties that help you to build attribution links, find copyright statements and, as a component of additional dataset metadata, publish structured citations.

Attribution should be simple and straight-forward. Even when a data license does not require attribution, you should attempt to acknowledge your sources.

The simplest form of attribution is to include a simple link in your application. For example:

<span>Uses data supplied by <a href="">Example Company</a>.</span>

The following steps suggest a simple algorithm for building an attribution link:

  1. Find the text to use as the anchor of the link:
    1. If the publisher has provided an odrs:attributionText property then use its value, otherwise
    2. If the dataset references a publisher via a dct:publisher property then use the foaf:name of the publisher
    3. If the dataset has a title property (dct:title) then use that
  2. Find the URL to use as the target of the link:
    1. If the publisher has provided an odrs:attributionURL property then use that URI, otherwise
    2. If the dataset references a publisher via a dct:publisher property then use its foaf:homepage property
    3. If the dataset has a dcat:landingPage property then use that, else
    4. Use the URI of the dataset

The approach given here is to use the preferred attribution text and URL provided by the publisher but fall back to using other dataset metadata if that is not available. A publisher may wish to direct attribution links to a particular target page to help track usage of their data (to justify ongoing publishing efforts) or to support attribution and acknowledgement of their sources.

While the example properties used in the algorithm draw heavily on DCAT, it should be possible to customise the behaviour depending on the data format(s) used by the publisher.

Displaying Copyright Notices

Some licenses may require you to display or preserve copyright notices that are provided by data publishers. For example you may need to retain notices when re-distributing data.

A copyright notice is defined using the copyrightNotice property, and will typically be a short phrase, e.g. © Crown copyright 2013, or Copyright © 2013, Example Ltd

You should also consider displaying any copyright notices found in a rights statement. A simple approach is to include the notice alongside your attribution text and link, e.g.:

<span>Uses data from <a href="">Example Ltd</a>. © Crown copyright 2013 Example, Ltd.</span>

If the data publisher has included a copyrightStatement link, but has not included any notice text, then you can build a link as follows:

<span>Uses data from <a href="">Example Ltd</a>. 
    <a href="">Copyright Statement</a>.</span>

See below for further notes on displaying attribution links and copyright notices.

Attribution in Linked Data Applications

If you are building an application that is consuming and displaying Linked Data, then you should also consider providing direct links to the resources that are referenced in your application. E.g. a community traffic monitoring application that uses the Ordnance Survey Linked Data might link directly to individual geographic areas (e.g. Bath) that are referenced from collected traffic reports.

This kind of deep-linking can help highlight which parts of a dataset you are drawing on. This is a more focused type of attribution enabled by Linked Data.

However this form of attribution is not always feasible. For example if a data view draws from on dozens of different Linked Data sources. For this reason the recommendation is to always include a basic attribution link and add “deep links” to Linked Data application on a best effort basis.

Displaying Links to Users

Attribution links should be displayed as close as possible to where the data is being displayed to a user. However the needs of your application and the limitations of the devices on which it is being used will impact your ability to prominently display attribution links.

As general guidance:

  • If your application uses a single primary dataset across the whole application, then include an attribution link in the footer of the application, along with any copyright notice provided by the publisher.
  • If your application uses several datasets, then consider displaying an attribution link next to a logical grouping of data items. For example if you display an “infobox” that provides some additional metadata, then include an attribution link in the footer of the infobox. Copyright notices might still be included in the page footer
  • If your application draws on many different datasets then add a colophon page that attributes each of the datasets in turn, including copyright notices.
  • If your application consumes multiple datasets but doesn’t display the results directly to users then you should still include attribution, again on a colophon or “about” page
  • If you are redistributing data, then ensure that copyright notices provided by the publisher are preserved, e.g. in a COPYRIGHT file in your dataset distribution.

Even when your application only uses one or a few datasets you should consider adding a colophon page that provides additional detail on the datasets you’re using and how you’re using them. This page makes a good location to include copyright notices and perhaps a dataset citation that can clearly reference your data sources.

Citing Datasets

Citation is usually more detailed than attribution and is intended to help support your users in:

  • Understanding precisely which datasets you are using — you may be using and attributing several datasets from a single source
  • Obtaining those datasets for themselves — e.g. to recreate your analysis, build their own application, or to report issues with the data

Dataset citations should include an attribution link, as well as more detailed metadata including:

  • The title of the dataset and a link to its homepage
  • The version of the dataset being used. E.g. its dct:issued or dct:modified dates
  • The name of the publisher
  • A link to the specific distribution used and the date it was obtained, if the distribution isn’t versioned
  • References to the Rights Statement that applies to the re-use of the data

The precise format of the link will depend on the needs of your application and users, as well as the detailed metadata available for the dataset.