Google Summer Of Code

[Enrico Zini]

On Sat, Mar 17, 2007 at 12:24:55AM +0100, Erich Schubert wrote:

> Summer of Code is on again, and Debian is in.
> The deadline is really soon, so hurry up!
> 
> It would be nice to have some Debtags related projects in GSoC, and we
> need good ideas for those. So if you are interested in GSoC, please
> suggest some on the list, we'll then give you some initial feedback.
> GSoC is about you, about getting you involved in the community and
> such; so we won't be as much proposing projects, but want them to come
> from you. So if you've had some UI idea sitting in you head for some
> time, it's time to write it down and maybe do a mockup! Is there
> something cool you'd like to do with Debtags? Propose it!

I can contribute two wishes:

 * Automatic detection of structure in the tag data

In the web editing interface at http://debtags.alioth.debian.org you can
see suggestions, like "this package has tag X, and normally packages
that have tag X also have tag Y".

Some of these suggestions are autogenerated by some little experimental
code I wrote on a train that can see simple relationships like "90% of
packages that have tag X also have tag Y".  

I've found that feeding back to the taggers suggestions based on current
usage produces a positive reinforcement of the structure emerging in the
tag data.  This gives us both liberty in tagging and order in the tag
data.

My code can now only compute one kind of probabilistic rule: ">nn% of
packages having tag X also have tag Y".  It would be great to have a
tool that is smarter, and can compute more complex, still probabilistic,
rules like:

 - ">nn% of packages that have tag X and Y also have tag Z"
 - ">nn% of packages that have tag X or Y also have tag Z"
 - ">nn% of packages that have tag X also have tag Y or Z"
 - ">nn% of packages that have one of the tags in set X1..Xn also have
   one of the tags in set Y1..Yn"
 - whatever else can be computed


 * Advanced interfaces that make use of debtags

Over the years I've produced lots of code that does cool things with the
tags.  Recently me and Arnaud Fontaine tried creating an installer
interface[1] that makes use of all this cool technology.

This has various possibilities of being carried forward:

 - usability work studying how all this technology can be used
   effectively in practice
 - more research on how to cross popcon and debtags data[2] to give
   effective suggestions to users
 - implement more of the ideas at [3]

[1] http://blogs.hurdfr.org/arnau/?title=pmnew_gui_announcement
[2] http://people.debian.org/~enrico/popsuggest.php
[3] http://www.enricozini.org/2007/debtags/pmdraft


Ciao,

Enrico

-- 
GPG key: 1024D/797EBFAB 2000-12-05 Enrico Zini <enrico at debian.org>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 307 bytes
Desc: Digital signature
Url : http://lists.alioth.debian.org/pipermail/debtags-devel/attachments/20070317/bb2d76c7/attachment.pgp