Re: [Demexp-dev] Delegation and ocamlgraph: an example and a list of issues

[David MENTRE]

Hello Thomas,

Thomas Petazzoni <address@hidden> writes:

> First of all, I'm not sure, as a demexp user that I'd be able to tell
> whether I would like to prefer tag "A" over tag "B". As a citizen, I
> really don't know how I could make an absolute decision to tell whether
> "Ecology" is more important than "Transport" for example. So, the tag
> preference mechanism solves the algorithmic problem, but does it really
> solves the political decision problem behind it ?

You could also have a different point of view on the issue by
considering which of the two delegates (the one of "Ecology" and the one
of "Transport") you trust the most for the questions having both
tags. Would you find the decision easier to take? 

Of course, taking a decision is blured by the transitivity of
delegation. 

> On a more general point of view, I think that the biggest issue the
> demexp experience will face is the "human scalibility" problem.

I heartily agree.

> The delegation mechanism system is supposed to solve the human
> scalibility problem generated by the number of questions in the
> base. However, this pushes back the problem to the number of tags, the
> management of the delegations, and possibly the management of the tag
> preferences (to solve the delegation conflicts discussed
> previously). Even if the tag set is supposed to be far smaller than
> the question set, I think it'll still be big enough to be hardly
> managed by an human person.

And hope that the meaning of tags is obvious enough! I agree that we
have pushed back a lot of issues into the classification system.

Keeping the number of tags under a reasonnable number is probably the
most obvious thing to do. For example, in Gulliver's experiment of
demexp, Frédéric has created a lot of tags. I think the classifiers
should agree on a basic classification, only creating new tags when only
absolutely necessary. Orthogonality of tags is also important. We should
probably think about a "code of tagging". 

But maybe we should think about the structure of tagging related to the
real world. For example, if you have a tag for each city in the world,
you are probably only interested to see, at first, tags related to
Toulouse and France, maybe not Doha or Canberra.

Regarding management of preferences, we (Frédéric, Félix and myself)
assume a transitivity of preferences. So if you prefer "A" over "B" and
"B" over "C", then you prefer "A" over "C". You should also consider
that conflicts are solved only when they arise in a lazy way. I _hope_
it will reduce the number of conflicts to solve after a cold start
period.

Regarding management of delegation, I have no real answer to
provide. But maybe the most sensible answer to give is to admit that
most people won't vote on the majority of questions, because they are
not concerned (e.g. Doha). 

Overall, I think we lack the "experiment" phase of demexp. I have not
used demexp much to vote and navigate. So it is difficult to tell if the
issues you underline are real issues or not.

Yours,
d.
-- 
pub  1024D/A3AD7A2A 2004-10-03 David MENTRE <address@hidden>
 5996 CC46 4612 9CA4 3562  D7AC 6C67 9E96 A3AD 7A2A