Re: Hierarchical tags

From: David Karger <>
Date: Tue, 15 Nov 2005 11:17:08 -0500

Drawing from a previous paper, I will note that there is no need for a
hierarchy of tags in order to offer the ui you are proposing. As with
our multiple categorization paper, I think it makes a lot of sense for
the next step (completions off "firefox |" to simply offer for
completion all those tags that co-tag and object tagged by firefox.
Hierarchies are very seductive but you tend to want different
hierarchies at different times, so it is bad to lock yourself in to one
when you could instead get the same thing by more of a "faceting" approach.

David Huynh wrote:

> I think providing a way to browse data hierarchically through
> user-assigned tags is the first item of business. This will look like
> a normal folder tree view.
> We can also provide tag completion not just at the time of tag
> assignment but also at the time of recalling items using tags. So, I
> can start typing a tag
> f
> and PB would complete
> f|
> irefox (39)
> riend (48)
> then I type i and hit tab
> firefox
> now when I type comma and space, I get other tags of things tagged as
> firefox
> firefox, |
> extension (18)
> flaws (6)
> hacks (13)
> javascript (2)
> David
> Stefano Mazzocchi wrote:
>> Brett Zamir wrote:
>>> And how about a browseable hierarchy for tags as well, drawing from
>>> Princeton's WordNet perhaps (as one option for those who didn't want to
>>> start from scratch), in order to allow the user to come across
>>> results for
>>> similar concepts (e.g., while browsing for the tag "pizza", one
>>> could easily
>>> get to tags for "pasta", etc.). This hierarchy could be editable
>>> along the
>>> lines of (and even interfaceable with) Wikipedia's category structure.
>> I've been thinking long and hard about this. Probably even too much.
>> Clearly, the lack of relationships between tags is damaging,
>> especially when the tag space starts to grow.
>> The problem, though, is that the relationships between those tags
>> should be folksonomized themselves, or we can run into trouble on a
>> globally distribute tag space.
>> For example: suppose you tag something with "blog" and I tag the same
>> thing with "blogs". The edit distance between the label of our two
>> tags is small enough that I can run a levenshtein distance and
>> present you (or me or somebody else) with a potential relationship field
>> blogs --- [edit text here ]---> blog
>> then you can type "plural of" and it becomes another statement. At
>> that point, now we have a relationship between your tag and mine,
>> clearly typed as a relationship. Later somebody else might say
>> "plural of" -(is a)-> "collapsable property"
>> so that the system might know that when you find blogs or blog they
>> really mean the same thing. (in OWL terms, a "collapsable property"
>> does the same of an OWL equivalence but in the folksological space)
>> The same emergence can be done by running such a distance against
>> wordnet, which would yield the ability to draw equivalences between
>> tags and words in wordnet... but again, those relationships are
>> *YOURS*, not global, then I can decide whether or not I agree or
>> disagree or even slightly want to differentiate myself from your
>> vision... defaulting to agreeing which is by far the most common case
>> in folksological spaces.
>> I'm not inclined to introduce something that can't scale to an entire
>> world of people tagging... and not tagging things apple or blog, but
>> things like abortion or war or holocaust or religion or terrorism,
>> where meaning *and* relationships are highly dependent on the
>> personal context.
>> And yes, there is an entire virgin research field on its own on doing
>> non-DL reasoning on a folksological space and I am very interested in
>> that research area myself, but tying tags to wordnet won't scale
>> socially, therefore has very little appeal to me even if immediately
>> would be useful.
>> That said, this is only my personal opinion and I don't speak for the
>> entire group... and remember, this is an open source project so
>> patches are always welcome :-)
Received on Tue Nov 15 2005 - 16:11:19 EST

This archive was generated by hypermail 2.3.0 : Thu Aug 09 2012 - 16:39:18 EDT