Re: [Nautilus-list] Fwd: Fer de Lance, intelligence for the free desktop

["Blad, John Erling" <john erling blad aftenposten no>]

Blad, John Erling wrote:
>..and to bring it a little back on topic for this list, an
>interface could be like an item in a context menu. Select a
>file then chose "Similar files" and you get a virtual folder.
>In this you can again choose "Similar files".
>
>For each sted the system learns more about what you are searching
>for and 2-3 step you should have a limited the querry enough
>for even very large datasets.

Ben Ford wrote:
>That is very slick.  You could also select multiple files to build the 
>query on.

Yeah but it wouldn't speed up the inference very much. Perhaps 3 steps
would reduce to 2 steps *sometimes*..

A real problem with the whole idea is that it doesn't integrate very
well with the present search interface.

For example if you do a qerry for a file and then choose similar files
on this file, should you then limit this search on your previous querry?
Should this be an illegal choice? Should you popup a dialog asking if
the search should be limited given the earlyer querry?

Even if Viper and Autonomy and systems like those are very fascinating
they can not say anything about what they find. In our therms here at
the newspaper we would say they lack meta information. For example Viper
can easilly find images which shows architecture but were are the buildings?
Who built them? When?

To be really usefull such a system must be integrated with other systems
that have this meta information. For example there are databases with
information about music, films, broadcasts, etc. Locally you have mime
types and other metainformation. In jpg files there are implicitt
metainformation. Rpm files also contains metainformation. Nautilus itself
might even add metainformation about files and directories.

John