SV: [Usability] User problems and practices with modern desktopsystems

[<john erling blad aftenposten no>]

I'll try to stick in some notes on your mail..

I'm not quite sure but it seems like I'm focus on the retrieval
situation and how to find specific stuff while you focus on how to add
such meta information? Could be we argue from different mental models.

I'm also a bit unsure about the actual words to use while describing
this as there are a whole lot of special descriptions which does not
translate to well. The most notable (if I'm not completely out on a
limbo):
Index			- list of descriptive words
Categories		- actual words (indices) in an index (metadata)
Classification	- assigning categories to an item (file, article, photo)

> >> What I would like to have or like to design is a new filesystem,
maybe
> >> based on a relational database, but not on top of an existent
> >> filesystem. The system may still run on a filesystem, but not the
> >> userdata, that what we now call "desktop" and the objects inside.
> > I would prefer if categorization and other indices about file
content
> > was stored as file metadata, sill the metadata could be indexed by
some
> > external means and stored in some relational database.
> 
> the metadata information must be stored in the "filesystem" because it
> should be copied or moved with your file.
> try to move data with an emblem in nautilus, the emblem is lost. i
> would expect that such information even belongs to a file if its moved
> by shell command, or copied via rsync for backup reason.
> With a flexible kind of metadata it may also be possible add
> information if the file was changed after copied. i often got the
> problem to find out what is never, the file at home or at work. the
> same problems appear with bookmarks, (yellow) notes, adresses etc.
> There are lots of solutions, but they individually depend on your
> software.

What I was thinking of was to use the existing metadata solution which
is an integral part of several file systems. This could be used as the
data storage mechanism while actual indexing could reside in any kind of
database, but still the original classification data is an integral part
of the file itself.

If the underlaying system supports copying files *with* metadata, then
they will not get lost. This way you could copy across systems as long
as the metadata is copied together with the file itself.

Such metadata should be fairly easy to index by most of the existing or
upcoming desktop search engines.

> >>>> This problem can be and should be tackled from both directions,
> >>> although
> >>>> automatically aggregating and searching information is useful it
is
> >>> also a
> >>>> good idea to encourage users to organise their data little
better.
> >>>
> >>> I don't think it is a very good idea to enforce to much manual
> >>> classification on the user. He simply won't use it. What I do
> 
> i believe that the classification now is to complicated and to
> restricted. if there is no folder that fit your classification you
have
> to
>   - open browse for other folders
> 	here you get struck with a lot of windows and functions (no
critic
> on
> the gnome file chooser, it is good, the 	problem is the
underlaying
> structure)

I guess this was an example of how you could use metaphores the user are
familiar with,.. Still I'm very aware that the current spatial nautilus
sucks for navigating unknown areas of the file system.

But the most important as I see it is that "mounted active folders"
would work no matter what application the user opens. Even old
applications would work, but note, this is for the retrieval situation.
Storage is something different and it is cumbersome to use this idea for
more than one classification. If you need multiple classifications you
must open a separate dialog somehow.

>   - hit create folder
>   - type the name
> 
> i prefer a text field one or more dropdown menues like:
> http://wkm.kunst.uni-wuppertal.de/~mueller/file.gif
> this is simply a first idea.
> 
> choose one or more categories or type an other topic. the categories
in
> the popup are the 10 most used in the past, and the text field uses
> autocomplete.

The main problem with this approach is that it is simply to difficult to
code such a solution for all the various classification schemes. For
example, if you compare classification schemes in newspapers you should
assume that they use the same classification scheme. This is simply not
so at all. There exist some standardized classification schemes but even
those are not comparable.

In general, coding for specific classification schemes does not work out
very well.

If I should try to follow your idea I would have a drop down box
labelled "Scheme" which picks from one of several classification
schemes. It should be up to the user which schemes he wants to use but a
reasonable set should be provided as default.

If a file has set a classification scheme then this is choosen as
default. There should also exist ways to coerce one classification
scheme into another without to much loss of data.

When one such classification scheme is choosen fields according to this
scheme are added to the dialog. I'm tempted to think you even need some
kind of subdivision of those as there can be a lot of fields. The fields
could be implemented as one of a limited set of input fields, but very
often you would want some kind of combo box with some kind of dynamic
code to automate retrieval of information from the file itself. 

To make thing even more complicated an item can use several
classification schemes. If you try to dismantle information from a
jpg-picture you could find information about the camera, photoshop and
all kind of stuff.

If you slightly extend the drop-down list you could imagine that this is
a list of *all* classification schemes which only a few actually
contains data. Then you could jump back and forth on metadata for a
picture, one time looking at camera data, the next photoshop data.

> in the filechooser there is the same field to blend all other files
> out, that don't belong to that category.
> 
> >
> > Search isn't very user friendly unless you are very familiar with
it.
> > My
> > idea of a balancing algorithm is something you could manipulate.
Let's
> > say you have a music folder. When this grows large enough
information
> > is
> > extracted from metadata to build an artist level. When you traverse
> > into
> > a folder with enough files this will again be divided into albums.
> >
> 
> i don't like search either. i prefer a situation that only shows the
> files that are important for me now and a set of tools that hide or
> unhide necessary files.
> http://wkm.kunst.uni-wuppertal.de/~mueller/browser.gif
> (some filter settings maybe possible today)

Here we somehow diverge. I would like to list files with specific
metadata content as if it were stored in a folder. You could imagine
this as combining metadata to extract files through a search. Something
like choosing music and Dylan. Wherever it is stored on the computer
shouldn't really matter. The actual "hits" could then be organised by
using metadata as sort criteria.

An alternate mental model is to have a special device which lists all
files available to the user, then you limits this complete set by
criterias on the metadata.

Now, you *should* be able to limit this set by adding free text search
or other kind of additional select criteria. Beat measures on music
files for example. And image classification algorithms so you can search
for similar files for example.

There was a system which had some kind of extension framework some time
ago. I belive the frontend was called snake charmer but can't recollect
any more about this system.

> > You could imagine that you was able to set one of several schemes
for
> > how this would be done. In fact given a specific type of folder a
> > scheme
> > could be predefined. A folder like "My Pictures" cold be balanced by
> > information from exif data.
> 
> exif is a good thing, but only for pictures, mp3 tags are only for
mp3.
> and both are different to read or change.
> 
> The desktop developement is going to be finished some day. The next
> step will be how we manage our data device independent and across the
> devices. thats needs a flexible and application independent set of
> information about our data.

Storage and retrieval of classification data should be more or less
orthogonal to how you organize your devices. Files/items that can't be
directly accessed should be shown as such with no other distinction. I
can't really see a reason why you shouldn't be able to search for Dylan
and get a lists of CD's where he performed.