Re: [gentoo-kbase] A Gentoo Knowledge Base

[R.I.P.]

Hi Sven, and all the interested people.

I'd suggest to have possiblity of several solutions of the problem.
Sometimes there is no good solution for the problem, but the number of dirty
hacks with advantages and disadvantages in each.
Another situation is when there is a problem  with a certain version of
software and there is two possible ways to solve it:
1. upgrade to possibly unstable but newer version in which the problem
doesn't exist
2. Fix the problem
In some cases 1st variant will be suitable, in others 2nd.

On 5/19/06, Sven Vermeulen <swift@g.o> wrote:
>
> I believe that some interesting natural language engines exist already
> which
> might make a good candidate. One of the projects that I have investigated
> thus far is the MindMeld project, which has a nice engine, giving good
> results for both "how do I" questions as plain copy/paste of error
> messages.
> It actually builds the result based on both its database and prior queries
> (with feedback from the user ala "was this answer helpful").


I've read about that MindMeld project. Its principle quite simple.
I doubt that MindMeld's engine's data will be poisoned with common words
specific to gentoo like "emerge","package" etc.They will be either cleared
by statistical filter or overwiegh other keywords.

From what I've read on MindMeld's site I've understood that they are using
some kind of 1-layer perceptron model. This model is well-known and highely
researched however in most cases it is too simple to consrtuct reasonable
neural networks for human language analysis.

And some IMHO about neural networks:
I have to say that I'm quite familiar with theory and practice of wokring
with neural networks.Their behaviour isn't very predictable. There is an old
principle: whenever you can avoid using of neural networks, you should avoid
it.
I'd advice to look towards some other engines which doesn't deal with neural
networks.

Categories are quite difficult to use in a knowledge base. A more active
> approach here is the use of keywords. If a topic has something to do with
> the installation, the keyword "installation" is active.
>
> Of course, viewing all topics with a certain keyword can be seen as that
> specific keyword categorie. All in the point of viewing.


As I've mentioned in topic on gentoo forums, I'd like to see cathegorized
keywords, possibly organised in some tree. Possibly there could exist 2
methods of searching: "by keyword" and "by natural language queries" or some
mix of both of them (Possibly searching with "natural language query" and
then filtering with keywords.
Organising tree of keywords is not that hard as organizing topics itself.
Even if you won't introduce full-featured keyword search it would be great
to separate by keywords different types of problems, like "configuration",
"install", "crash", "dependency" problems etc. This could be applied like an
additional filter to the query results.


The Gentoo Documentation Project should focus one other things, like general
> guides, internationalisation of the available documentation, search
> ability,
> published articles, project documentation and the lot.


Also one more thing came to my mind. Do you plan to allow
internationalization for the KB project? If so, there is no point in
sticking to some "natural language engines" that doesn't allow extension to
the other languages.

---------
Ты тонкий мыслитель сказал людоед, а больше толстых люблю (с) me