[Tagdb] Fwd: [delicious-discuss] tag pecularities in a search engine

[Brian Del Vecchio]

Amir Michail (of CollabRank) wrote an excellent comment on the
delicious-discuss list which I thought would be well received here.

Apologies to subscribers of both lists who get two copies of this.

---------- Forwarded message ----------
From: Brian Del Vecchio <hybernaut at gmail.com>
Date: Aug 19, 2005 12:32 AM
Subject: Re: [delicious-discuss] tag pecularities in a search engine
To: Amir Michail <amichail at gmail.com>
Cc: discuss at del.icio.us


Yes, that makes a lot of sense.   The recommendation engine announced last week
  http://lists.del.icio.us/pipermail/discuss/2005-August/003749.html
is similar, but I think it's geared more towards discovery than
search.  One could easily imagine an "expand this search to include
similar tags" flag on search.

However, if you're using the new broad search as in
  http://del.icio.us/search/all?search=movie%20reviews
then one of the effects of the broad folksonomy (as Vanderwal puts it)
is that the users already do that for you:  most of the items tagged
movie are also tagged movies, so it doesn't matter which variant you
search for.  No extra work is required.  Of course that works better
for items with many tags.

Also, this is one of the reasons why Joshua (and others) resist making
it easy to copy tags when you're copying a post.  Manually entered
tags are far more valuable, in part because of the variations.

PS Amir, in case you haven't already seen this, Nitin Borwankar has
started a mailing list
  http://lists.tagschema.com/mailman/listinfo/tagdb
where this topic would be well received (Not that it is unwelcome
here).  I'll cross post this to tagdb in the hopes of starting a good
discussion over there.

On 8/18/05, Amir Michail <amichail at gmail.com> wrote:
> Hi,
>
> It seems that searching using tags is a bit tricky and can give
> unexpected results.
>
> For example, one might expect the query "movie reviews" to work well,
> but perhaps "movies reviews" works better (e.g., in cases where most
> people use the tags movies and reviews rather than movie and reviews).
>   Tags associated with a URL don't need to form a phrase after all.
>
> Of course, there are other problems.  For example, variations such as
> "computer science", "computer_science", "computerscience", "compsci",
> and "cs".
>
> It seems that a search engine based on tags needs to handle these
> cases in a way so that the search engine user need not be aware of
> these variations.  Perhaps, any variation should work as a query?
>
> So the query "movie reviews" should work well even if most users use
> the tag "movies" and any variation of the cs queries should work as
> well.
>
> This can probably be done with some data mining.
>
> Amir
> _______________________________________________
> discuss mailing list
> discuss at del.icio.us
> http://lists.del.icio.us/cgi-bin/mailman/listinfo/discuss
>

Brian Del Vecchio   |   bdv at hybernaut.com   |   http://hybernaut.com/