Re: "Tag" metaphor

> Hello all,

Hi Ben,

> With the recent tag ontology RFC and official announcement of 
> "tagtriples," what are people's thoughts as to how tags fit into the 
> goals of the Semantic Web? When I follow the "tag" metaphor, I think 
> of a tiny bit of shorthand information temporarily elastic-banded to 
> something else. The tag often has meaning in its specific context; in 
> a furniture warehouse, a tag with four letters might be an inventory 
> code. Yet a nearly identical furniture warehouse just across town 
> might use a different annotation to mean the same thing. Furthermore, 
> the four letters might have a separate meaning to someone else (such 
> as if they spell out "DUCK"). Who do you trust? The majority? How is 
> this the tag metaphor something to aspire to?

I quite agree. Typically I see them as being shorthand for more 
thorough semantics. When using iPhoto, I rely on its tag support for 
describing things because it lacks better descriptive elements, though 
I have to be careful of overlap (the context problem). Tagging has 
three advantages:

- It's quicker to tag something with "Highlander" than to apply the 
complex chain of RDF "this photo depicts an event that took place in 
the Highlander pub in Windsor";
- It's straightforward to browse/search etc. on tags, as long as I can 
remember (or synthesise) the tag that I would have used;
- The interface for tagging is simpler than for more advanced 
operations (Tidepool excepted, of course!).

For most applications I think it's this apparent simplicity that makes 
tags worthwhile; the interface to del.icio.us or iPhoto is simpler 
because the complex semantics of each tag resides in the user's head. 
It's also handy that typically someone else's "Python" tag applies to 
mine, though "Highlander" is quite the opposite!

> 	Tags seem very interesting for organizing bookmarks and grouping 
> photos, but will you ever be able to carry out any inference with 
> them? Because the tags themselves are ambiguous, you wind up with all 
> the old problems of distinguishing "python" from "Python," etc. Add to 
> that the issue of millions of users creating tags, and you have a 
> nightmare of overlap and duplication. What can be done with the triple 
> "Becca plays JazzFlute," by a non-human? The reasoner can then see 
> that someone has said "JazzFlute type MusicalStyle," but someone else 
> has said "JazzFlute is SceneInTheMovieAnchorman."

Yes, there are massive problems with tags. However, they have their 
advantages; a lower barrier to entry is one, mapping directly to 
complex concepts is another (e.g. I wouldn't actually know which 
vocabularies to use to do my pub annotation!), and aggregating 
disparate resources is another (e.g. "DSLs" applies usefully to 
Haskell, Graphviz, papers on DSLs, and a whole host of things that are 
indirectly related to one another, and I might want to find any of 
these things).

> 	The latter statement would be increasingly more likely if we were to 
> stick to strict triples, as in tagtriples. How would you say "there is 
> a scene in the movie Anchorman that depicts Jazz flute" without 
> Bnodes?

With great difficulty, I would say --- but just building the annotation 
in RDF would be hard enough. The obvious response is "which is better? 
having [JazzFlute appearsIn Anchorman], with the associated wishy-washy 
semantics, or no annotation at all because the thorough version is too 
hard to construct?".

In other words, I think something like tagtriples has its place, 
particularly for end-user annotation.

> I'm not trying to be contrary; I'm just curious, as a newbie, what the 
> advocates of tags have in mind for them in a broad SW context.

I'll take this opportunity to state that I'm not an advocate of tags as 
such, but I do see their benefits. I think that where tools provide 
sufficient support, "real" meta-data are much more useful. Where users 
are throwing together descriptions on the fly, however, tags are much 
easier to use. It's the current folksonomy vibe. Danny's use of 
WordPress's categories to generate SKOS Concepts is I think the best 
blending I've seen.

Two of the things I'm interested in, with my iPhoto RDF plugin (must 
update it to work with iPhoto 5) and also with my current work, is (a) 
mapping tags into RDF, and (b) relating tags to one another. I think 
they might be fruitful areas to explore.

-R

Received on Wednesday, 23 March 2005 17:13:13 UTC