[Date Prev][Date Next][Thread Prev][Thread Next][Author Index][Date Index][Thread Index]

Re: Fuzzy Hypertext?

At 12:26 PM 12/19/96 +0000, Paul Newton wrote:
>Anyway - some hypertext theory! I have recently been applying the idea of
>fuzzy logic to hypertext. I find that current implementations of hypertext
>can be very restrictive by forcing the reader/author to work with very
>static and discrete linking models. I realise that there has been much work
>on filtering of links and more dynamic systems, but I would like to see
>some representation of the fuzziness that must surely exist when an author
>creates a link, or a reader follows a link.
>When an author creates a link, I would like them to be able to give the
>link a fuzzy level, so that they can express uncertainty eg links which are
>definite, sometimes useful, only useful to the dedicated reader
>etc. Further, I find it very restricting that anchors exist as such
>discrete objects. It can be quite natural to link specific words
>(dictionaries, indexes, tables of contents etc), but there must be cases
>when hypertext would benefit from a more fuzzy definition of anchors. For
>example, when I discuss a concept, I can rarely identify precise points
>where the discussion starts then ends - there is a build up to the ideas,
>then a discussion of consequences etc. This is a much more fuzzy definition
>of an anchor, where instead of discrete boundaries, we instead have more
>fuzzy, blurred edges.

This idea has an additional benefit, especially in the case of hyper-linking
models which seperate the link declaration out of the context of and of it's
anchors (a-la HyTime).  You could specify that you are only interested in
links which exceed a specific level of certainly (the fuzzy logic you are
talking about is as much certainly modelling, which predates fuzzy logic, in
that you have not specified where the "logic" in fuzzy logic is being applied).

That idea actually makes me think this might be better fitted to placing
links in a N-dimensional space rather than a 1 dimensional space, as
described.  This is moving toward a more neural-network style model.  Each
dimension could either explicitly or not, model some information about the
link.  It would then be possible to ask for link which are similarly typed
to some set of links you already have.  Or if you the dimensions correspond
to exact typing/style info, you could say get me all the links which a 90%
linear-traversal, or 70% definition....

The problem with a simple certainly model as Paul described in his original
post, is that it can only ever model one type of information about the link.
If that is all you want/need, then it is perfect (and much easier)... The
other issue is thatof scale.  One author's 90% is not necesarily equivalent
to another author's 90%. (stupid human tricks...this is part of why most
rating systems have explicit, carefully defined, limited states.)

"that which is not slightly distorted lacks sensible appeal: from which it
 that irregularity - that is to say, the unexpected, surprise, and astonishment,
    are an essential part and characteristic of beauty" - Charles Baudelaire