Understanding metadata

What metadata is and how to use it to check if a dataset is right for you.

Metadata is information about a dataset. It includes:

  • the title

  • what the dataset covers

  • who published it

  • when it was last updated

You can find metadata on a dataset page alongside the data files.

Why metadata matters

Metadata can tell you if a dataset is:

  • relevant – does it cover what you need

  • up to date – when it was last updated

  • trustworthy – who published it and under what licence

  • usable – the format, time period and geographic area

Checking metadata

To help you decide if a dataset is right for you, check the following metadata fields:

  • title and description – what the dataset contains in plain English

  • organisation – the public body that published it

  • published – when it was first published

  • last updated – when the data was most recently refreshed

  • next update – when the next release is expected

  • update frequency – how often new data is added

  • reference period – the time period the data covers

  • geographic coverage – the area the data covers

  • licence – what you can do with the data

  • contact – the email address of the data publisher

Example

If you wanted to find data on recent hospital admissions for NHS Lothian, you could check the metadata of a dataset to confirm:

  • who published the data

  • it covers the NHS Lothian health board area

  • it was updated this quarter

  • it's published under an open licence

Metadata formats

You can download metadata in machine-readable formats. This is useful for developers and analysts building integrations or analysing the data catalogue.

We publish metadata using three vocabularies in a few common serialisations. A ‘vocabulary’ defines what each field means. A ‘serialisation’ is the file format the vocabulary uses.

DCAT-AP 2 (data catalogue vocabulary application profile)

This is a vocabulary for describing public-sector open data. It's used by open data sites across Europe. We publish DCAT-AP 2 metadata in RDF/XML, JSON-LD and Turtle formats.

Find out more about the DCAT-AP 2 vocabulary on the GitHub website.

Schema.org

This is a vocabulary for describing structured data on the web. Search engines use it to read and display dataset information. We publish Schema.org metadata in JSON-LD format.

Find out more about the Schema.org vocabulary on schema.org.

Croissant

Croissant is a vocabulary for machine-learning datasets, built using Schema.org. It adds extra details that machine-learning projects need. This includes information about data structure and how datasets are split for training and evaluation.

We publish Croissant metadata in JSON-LD format.

Find out more about the Croissant vocabulary on the ML Commons website.