beyond text Re: Attaching Meta-Data

[Srikant Jakilinki <sriks dcs gla ac uk>]

Hi Julian and Joe,

Your document indexer Julian? Is it a patch you wrote for Beagle or do
you have your own software? If so, would love to have it...

Re-engineering application approaches. Hmmm... it is easy for Apple to
do such a thing but for us it is basically loss of freedom. I do not use
Ximian Evolution nor GAIM but something else and when Dashboard only
works for those, I will be very sad. I think we need a VM kind of thing
that works at a granularity of files before they get sucked by
applications into their own native formats... I think all the context
you mentioned Joe is "locked" within the applications. They should
"release" this metadata. Cheers for the example and scenario. Long way
to go before we start to think of such maverick ideas I suppose :-)

Yes - the mention of RDBMS is especially relevant here. I wrote a
section in my first year PhD mini-project on M-Rules (short for Metadata
Rules) and the example I gave was that of how one can use EXIF values to
categorize digital images. Identifying patterns is obvious but I wrote
about a community based approach to categorize information...

For example, a photographer among us can write a rule called "outdoors
or scenery" which is essentially EXIF values of -
Size >= 800x600
Flash = Off
Exposure = ?
The indexer could then attach-the-metadata of "outdoors or scenery" when
it parses the EXIF information of photos and gets a match based on this
rule. I may be talking over my head here but I think indexing should be
a 2-way process...

Of course such things need not apply to photos only. When we use clipart
(and the indexer knows what clipart has been used in a document by
parsing the office formats), then we can attach-the-metadata that the
clipart has (like say, "idea" or "mobile-phone") to the document. I
think it is called attribute-transfer or something. Will check...

I am sorry guys if all this is a bit vague but I am very interested in
Dashboard/Beagle pushing the boundaries of indexing/searching with novel
approaches rather than doing the regular stuff of writing filters for
every known file format etc. Do not get me wrong but filters are
necessary but I would really like it if we could add more intelligence
like being able to extract images from PDF documents and then marking
all other PDF (or office) documents similar if the images extracted from
these are similar (contentwise) as well. For example, manuals or
articles having logos. I think we should start to look beyond text...

On Thu, 2004-10-21 at 09:33, Julian Satchell wrote:
> If it is any help, the last version of my document indexer had some
> initial support for meta-data. 
> 
> There are a number of subtle design issues for efficient retrieval on
> meta-data oriented searches; in a RDBMS design you tend to end up with
> many-way joins.
> 
> Julian
> 
> On Wed, 2004-10-20 at 15:37 -0400, Joe Shaw wrote:
> > Hi,
> > 
> > On Wed, 2004-10-20 at 20:33 +0100, Srikant Jakilinki wrote:
> > > Guys, any cool ideas and scenarios that you have regarding "attaching
> > > metadata"? I think we should probably focus on this aspect a bit more
> > > than we have done till now. 
> > 
> > I think the main thing at this point is that we're losing all kinds of
> > metadata when we interact with others.  When you save an email
> > attachment you lose all contextual information there.  Who sent it, when
> > did they send it, who else did they send it to, what other attachments
> > were sent, what message/thread was it from, etc.  There are similar
> > issues with instant messaging.  URLs and files are passed around there,
> > but there isn't any kind of link established, and so there's no useful
> > way to re-associate it with what I'm really looking for.
> > 
> > Identifying those situations and then addressing them (probably
> > instrumenting the apps somehow) is going to be one of the most
> > challenging tasks of the project, IMO, but also one of the most
> > rewarding.
> > 
> > Joe
> > 
> > 
> > _______________________________________________
> > Dashboard-hackers mailing list
> > Dashboard-hackers gnome org
> > http://mail.gnome.org/mailman/listinfo/dashboard-hackers
> > 
-- 
Cheers-Regards-Sincerely,
Srikant
"How does one identify subtlety?" - Sriksisms ~powered by~ TagZilla
http://sriks6711.blogspot.com