Baldo,

Nice question :-) which brings several ideas to mind. First, your last
comment seems most promising. Ideally I think you need a generative system
that will give rise to the space structure you want. The question is how? If
you look at the early chapters of the Social Logic of Space (Hillier &
Hanson 1984) you will see some simple generative models - rule restricted
random aggregations for certain kinds of 'beady ring' settlement. A couple
of years ago we developed these ideas as a method for laying out objects in
a virtual environment to represent documents in a shared workspace. The idea
was to develop a landscape that was 'meaningful' by adjusting the
aggregation probabilities such that clusters would emerge of objects
representing documents that were similar on some parameter - say key word or
author or date. We developed a number of different versions of these
aggregation rules, for instance ones in which the key criterion was
intervisibility through open space of 'similar' objects, or metric shortest
distance clustering, or in which clustering was on the basis of more than
one criterion, or where objects could be relocated to optimise the layout.
What we did not do, but should have, is to evaluate the intelligibility of
the spatial structure and optimise for that. The reason we did not do it was
that this would have greatly increased the computational time for each
iteration of the growth process.

Now if you were to aggregate on the basis of pagerank, but to require
maximum metric separation of high pagerank objects, followed by
intervisibility between them, that might give rise to something interesting.
Essentially, the first block to be placed in the landscape would be the top
pagerank page. The second would be placed a long way from it, the third a
long way from both (ie. they form a triangle) etc. but as you progressiviely
fill in with lower pagerank objects your rule would maintain intervisibility
between the high pagerank objects. What should emerge is an axial space
structure linking high page rank objects across the landscape. As the space
fills up lower pagerank objects will maintain visibility to this larger
scale grid, but there will be less opportunity for the intervisibility to be
maintained and a smaller scale axial structure ought to emerge - this might
give rise to the correlation between global integration and local axial
connectivity that characterises an intelligbible axial structure. This is
all a supposition, my experience is that these things never do what you
expect, but something along these lines might work...

I suspect the thing to do is set up an experimental aggregation system and
try it....

Good luck and keep us posted with the results :-)

Alan




Baldo Faieta Wrote:

Hi everyone,

I have been trying to do a parallel between space syntax techniques and
current
information retrieval techniques in order to build a navigable urban grid of
the whole web or at least of a portion of it.  There are various motivations
to
do this. In a link-based model, as the web is organized, there is very
little
sense of place because the users don't have a view of where they are.
Following link to link, users can get quickly lost, and when there is an
underlying organization (usually local to the website) to help them, it
tends
to be a hierarchy which has to be painfully learned. Also, browsing is slow
because you only make use of your cognitive abilities (not your perceptual
abilities) to figure out where to go.  Within a particular page there are
many
links where you can go next and when you go to the next page you have this
very
narrow view of where you came from or where you are going.

With a spatial organization, and in particular with a grid layout, I hope to
overcome some of these limitations.  In my view, the user would move down
corridors (streets?) where web pages would be tiled next to each other in
either side of the walls of the corridor.  Corridors would connect to other
corridors and presumably there would be enough navigational signs analogous
to
street signs in cities.  The placement of the web pages in the corridors
would
be such that if you look at all the trails that people take in the web,
pages
near each other in the grid would be, on average, close to each other in the
trails.

An implicit assumption with this view is that web pages are locations, yet a
lot of web pages would not be considered locations but more navigational
pages.
 Consider for example, the pages of the open directory, dmoz.  There, most
of
the pages are really there for navigation and to access other web pages.
So, I
would have to make the distinction between web pages that can be considered
as
locations or destinations and those that are navigational and these
navigational pages can be candidates for streets (the corridors).  There are
also groups of web pages (web sites really) linked to each other that
attract
users with similar life styles and these users could be considered a
community
around them.  In a form analogous to (some) real life streets, we should be
placing together all these pages in the same street so that you get synergy
between each other and support its corresponding community.

Assuming we have an "intelligible" grid layout, the user should be able to
move
quickly through the corridors and also make use of his/her perceptual
abilities
to look for information. Also, (again depends on a good layout), remembering
locations (pages) and learning a mental map becomes possible, making use of
our
abilities to navigate in our (physical) urban spaces.  This organization
might
even afford a real-time sense of place in much the same way as streets do in
that if the browsers (users) would see each other, presumably it would
increase
the chances for interacting with each other in a meaningful context.  This
representation, of course, would be as a complement on the current web
rather
than as a replacement and not every web page would have to be accounted for.

What I am working on is a way to use the latent information on the topology
of
the web to figure out a good mapping for a grid layout of all or a partial
set
of web pages.  If you consider how google works, they take the link
information
within web pages and out of this link topology they can effectively figure
out
which web pages would have the highest traffic.  Actually this is not quite
correct because they really calculate an authority measure (its pagerank),
but
there are some alternative techniques that do predict the traffic flow,
though
both measures are calculated with a similar effort.

When you do a search, they first figure out which pages have the keywords
you
put in (again this is a bit simplified) and then they rank them using their
corresponding (global) pagerank.  The pagerank measures, on average, the
proportion of random surfers (i.e., start from a random page, and then
select
one of the outlinks from the page in a random fashion) that you might find,
at
any one time, in a page.  In addition, out of the pagerank calculation, you
can
also derive an inverse distance measure between web pages (i.e., higher
values
correspond to pages close by in the link topology and close by means that if
you consider all the web trails, then they would be close to each other in
the
web trails that contain them).

In order to build the grid, I was making some parallels between web pages
and
locations in a grid.  A natural parallel might be that high pagerank pages
that
also work like hubs (they point to many other pages), would correspond to
high
integration streets or lines in the grid (using the space syntax meaning of
integration).  Presumably, they have to be long lines and be shallow in
terms
of accessibility.  As explained above, I have to decide when to place a page
in
a street and when to make it one.  I probably have to make sure that the
grid
is itself intelligible (using again the space syntax meaning).  I guess I
was
thinking that a good grid layout would be such that if you do a space syntax
analysis, I should find that some measures (integration, visibility, ?)
would
correlate well with their corresponding pagerank scores in that implicitly
they
might be measuring similar behaviors.

I was hoping that people in this list might provide me with pointers to
space
syntax literature I should look at and of issues that I should be aware of.
Some of the mappings that I don't know how to think of is when to decide to
link streets together and how to think of some of the ideas of the super
grid
(that is, different levels of traffic in a grid).  I am assuming that the
psychological principles behind the navigation in a physical urban grid
would
be similar to this generated grid and they should transfer over.  I guess
from
the limited knowledge of the space syntax techniques that I have, I haven't
found much in terms of generative principles.

Any help would be greatly appreciated.

Baldo