A dataset may have a number of different entities, some of which are simple, e.g. people or organizations, whereas others are more conceptual or complex, e.g. an observation made at a particular point in time, under specific conditions. It may not always be clear to a developer, or a user exploring a graph in a browser, what a particular resource represents

Ensure that every resource in a dataset has an rdfs:label property

   
ex:Book
  rdfs:label "War and Peace".
  
ex:WeatherObservation
  rdfs:label "Rainfall measurement from Weather Station 1 recorded by Bob on 17th August 2011"; 
  ex:rainfall 50;
  ex:date "2011-08-17"^^xsd:date
  ex:location ex:WeatherStation1;
  ex:experimenter ex:Bob.  
   
  

The rdfs:label property is a useful generic property for labelling any type of resource. By using this generic property to label any resource we can ensure that applications can easily discover a useful default label for a specific resource using a well-known property. This is particularly useful for supporting browsing of a dataset, as a browser can look for a default label. Developers can also use the label to assist in debugging queries or exploring a dataset.

Client applications may not always wish to use a provided label instead preferring to construct them based on other criteria. The Preferred Label pattern recommends using the skos:prefLabel property to communicate to clients a preferred label specified by the data publisher.

In some cases both a rdfs:label and a skos:prefLabel (or other specific labelling property such as dcterms:title) might be provided for the same resource. The content of the labels may differ reflecting the slightly different semantics of each property, e.g the rdfs:label might be longer or more descriptive than a shorter skos:prefLabel. If both label properties are provided with the same content, then this is an example of the Materialize Inference pattern: skos:prefLabel is a specialization of rdfs:label.

The importance of applying labels to Linked Data, as well as evidence for the poor adoption of this practice, is given in a paper called "Labels in the Web of Data" by Basil Ell, Denny Vrandečić, and Elena Simperl.