Notes on Modelling RDF Data
When modelling any data in RDF a number of decisions need to be made about how to construct the data.
The way data is modelled will affect:
- How easy it is to query in future;
- ?
Examples
Eurostat
Halb et. al (2008) explore the modelling of Eurostat data. They choose to model the data by:
- Items being contained within a Dataset; (Dataset items are related to their dataset by skos:narrower, from the widely used skos schema/ontology for describing 'Knowledge Organising Systems')
- Items containing a single value;
- Items having 'dimensions' such as the year or geographical areas they apply to;
They re-use a range of schemas including:
- Dublin Core (DC) Elements and Terms
- Geonames
- Simple Knowledge Organisation Systems (SKOS)
- Description of a Project (DOAP)
- the event ontology [16]
They contrast their modelling of data to modelling of US Census Data which uses complex compound properties such as "population15YearsAndOverWithIncomeIn1999". In this model you would need to find the name of the property "population15YearsAndOverWithIncomeIn1999" to query it - and the computer could not easily work out that this contains multiple facets (with or without income; age; year) that could be navigated. By representing Datasets, items and dimensions separately, Halb et. al's model attempts to be more flexible, and to allow new datasets to be added alongside the existing datasets without upsetting the data structure.
Links between data items and other online resources can be made in Halb et. al's demonstration using three predicates:
- rdfs:seeAlso
- owl:sameAs
- foaf:topic
Each gives the link a different meaning.