From Media Annotations Working Group Wiki
Jump to: navigation, search

This is a short review of Flickr's machine tags, introduced on the mailing list by Felix, and presented here:

What are machine tags

Basically, a machine tag is just a tag, but with a particular structure. More precisely, a machine tag has the form namespace:predicate=value. It has apparently been long used by flickr itself to encode geo-localization (geo:lon=123).

What has become special about machine tags is that they are now acknowledged in flickr's API (not on the website yet): one can query flickr for photos with machine tags, using wildcards in any of the three parts:

  • "dc:title=" to find any photo with a value for dc:title
  • "*:title=" to find any photo with a value for title in any namespace
  • "dc:*=\"champin\"" to find any photo with "champin" as a value of any machine tag in the dc namespace

Note that it is suggested that the value of machine tags can be structured as namespace:suffix as well, though the api does not provide means to use that.

How does it relate to RDF

Machine tags associated to a photo are much like RDF triples having that photo as its subject. The main difference is that the so-called "namespace" of the machine tag is not necessarily associated to a URI prefix. So there is no guarantee that different people will not use the same prefix with different meaning. But this was already true of plain tags, and is considered as a feature rather than a problem in the land of folksonomies :)

It is even suggested that one could use machine tags of the form xmlns:dc= if they really wanted to be unambiguous (although the API will not use it).

What interest for the WG

It is an interesting case of bottom-up emmergence of structure from a very simple conceptual model (tags).

If extensibility is considered a desirable feature of our ontology, this could help us to foresee how this extensibility could be achieved.

Finally, it could be interesting to investigate which are the machine tags actually used in flickr, in order to judge the relevance of metadata fields. TODO (pchampin).