Using UCDs as tags

Martin Hill @ ROE mch at roe.ac.uk
Wed Aug 13 04:23:16 PDT 2003


> At the contrary, and that's where I go extreme, I think that all metadata
> should be part of a well defined dictionary. Not only that, but even 
> metadata
> values should be part of the same dictionary!

Yes, we need such a dictionary if only because we need to be able to have 
our software compare like with like even if they look different (eg 
different intensity measurements) as you had at the end of your email.  
However this is also the core tricky bit to this approach!  As far as I can 
tell, all efforts so far to define a dictionary of complete and unique 
terms have failed because it's very difficult.  For the near term, as a 
temporary fix, we will be using UCDs but the UCD team are also not willing 
to take on the effort of making their dictionary unique or indeed complete.

However anyway I think it is easier from a coding point of view to 
categorise values in tags, rather than putting too much specialised meaning 
into the tag itself.  Using your example, compare:

<INSTRUMENT> WFPC2 </INSTRUMENT>
 with
 <WFPC2/>

It is easier to code a single *class* of 'Instrument', of which there will 
be various *instances* including WFPC2.  Otherwise we could end up writing 
a new class for each instrument.  We can also search through all our 
instruments looking for particular ones, and we can extend the tag to 
include more information, such as:

<Instrument>
   <name>WFPC2</name>
   <passband>
      <centre units="elephants">12</centre>
      <width units="elephants">1</width>
   </passband>
</Instrument>

The thing with us humans using the term 'WFPC2' is that we automatically 
expand it into the full meaning (or the relevent useful bits) in our heads - 
 but a computer needs to have all this meaning written down somewhere, and 
inside the tags seems to be the right place to have it.

I am however a great fan of using meaningful tags, and using UCDs as tags 
for data values sounds like a very interesting approach!

Cheers,

Martin

-- 
Martin Hill
Software Engineer, Astrogrid
07901 55 24 66



More information about the registry mailing list