Map of knowledge
at http://www.nytimes.com/2009/03/16/science/16visuals.html?_r=2&emc=eta1
<http://www.nytimes.com/2009/03/16/science/16visuals.html?_r=2&emc=eta1> built
by scientists from LANL, SFI etc.
I must admit, I have a hard time working out what these network
visualizations are meant to be telling me. That academic disciplines
are connected? Did I *really* not know that before looking at the
pretty picture?
You are not alone in this observation... but I, for one, do get a lot
out of it, and can imagine getting a lot more if I had direct access
to the data (and tools not unlike the one used to create this "map").
http://www.nytimes.com/imagepages/2009/03/13/science/16visual-popup.html
is the entire map, not a cropped subset.
1. Let me start with the disclaimer that I was in no way involved
in this work. There are others on the list who are at least
close to this work if not directly involved. I hope they will
chime in...
2. I don't think the goal of the project was to create this
particular visual. This particular visual (the whole one, not
the cropped one in the article) is surely used mostly
"iconically" to give the layman a sense of what the work is
about. Imagine if no such visual were included in the
article... even more opaque I think. A list of how many
articles and the major (conventional classifications they were
in) and the number of links between the classifications seems
like about as far as you could go, especially for a lay publication.
3. To whatever extent the researchers use visualizations like this
for Analysis, they probably use many... with different
thresholding criteria, different subsets, etc. I myself,
prefer a completely dynamic, interactive network layout for
analysis. In fact, I prefer one embedded in a 3D environment
which I can explore more directly.
4. In my work in SciViz, InfoViz, and Visual Analytics, I would
claim that virtually none of the visualizations my colleagues
use for doing analysis would be immediately useful to the casual
observer. Those which are not particularly abstract (fluid
flows) or very familiar (conventional charts and graphs) might
be recognizable, but not necessarily useful. How many people
would know to look for or recognize a "bowtie" in a
computational mesh? How many would see that the adaptive
meshing technique was failing in a region of high change? Etc.
Even simple charts and graphs intended for analytical use are
opaque to the layman. So, I can tell that the concentration of
a particular ion goes up roughly exponentially with one factor
and more linearly with another... so what?
5. Even Geography/Cartography can elicit a "so what"? There are
big deserts in along parts of the equator, rain forests along
other parts, I bet it is hot there. Mountains seem to come in
long skinny ranges or big clumps. Coastlines are ragged. The
names of countries in South America seem to be Spanish. There
are a lot of countries in Europe I never thought about because
they were formerly lumped in with the Soviet Union. Didn't I
know all those things before I looked at a world map with
geopolitical features marked? Actually, I probably learned them
from maps I have seen all my life.
6. In my experience, especially with Visual Analytics, the goal is
Exploration, Discovery and then maybe, sometimes Analysis.
Exploration and Discovery are a lot more "fun" even if the real
work is in the Analysis.
7. Network Science is not new, but it has only been about 10 years
that it has become highly popular and widely used. The visual
(and linguistic) idioms are still somewhat young and we haven't
all learned to read/think with them.
Going to the actual network diagram...
http://www.nytimes.com/imagepages/2009/03/13/science/16visual-popup.html
Without knowing the key to the node size and colors... I can intuit,
or extract some interesting (to me) things.
1. There are a few large clusters of relatively tightly coupled
subjects which are relatively distinct from eachother.
1. Soft Sciences, Religion, etc.
2. Biology, Environmental Science, Ecology, Agriculture
3. Hard Sciences, Physics, Chemistry, etc.
4. Health Sciences
2. The biggest "wad" are what some of us would call the "soft
sciences". It might not surprise some of us to notice that Law,
and Education and Philosophy are fairly entertwined. It *might*
surprise some of us that statistics is so connected.
3. there is another "big wad that we might generally refer to as
the hard sciences.
4. It might surprise some of us that Biology seems to be somewhat
distinct from the other sciences, connected through
biochemistry, toxicology and biotechnology.
5. It might inform, if not surprise some of us to realize that
Psychology might be tied to Biochemistry and that Biology ties
to Architecture and Design through Biodiversity and Ecology.
6. It surprises me that the wad on the left in Red which roughly
seems to relate to Medicine in general, doesn't tie in with
Physiology and Genetics from within the Biology Cluster.
7. Does it surprise us that Statistics is tied to Medicine through
Demographics and then through Clinical Trials?
It may be my experience and normal role, but an important thing I
think I see in this visualization is that either the data or the
tuning of the parameters might have artifacts. This Visualization
was probably not tuned for Analysis, or if it was, it was tuned for
one aspect of the data. It was probably tuned to make a pretty
picture so folks who know nothing about what they are doing, would at
least be able to see the rough structure and symmetry. No criticism
of their work here.
1. Why is Pharmaceutical research disconnected from Clinical
Pharmacology?
2. Cognitive Science and Neurology?
3. Where is Engineering?
4. Why is Tourism there?
5. What else is missing, obscured, or that I'm not noticing?
I immediately want to do several things:
1. move this into 3D so there is more "conceptual layout space" and
so I can adjust perspectives to see different otherwise occluded
features.
2. make it dynamic so that I can "pluck" portions of it and watch
the disturbances propagate, adjust parameters and watch it evolve.
3. play with the parameters to accentuate tight clusters or lightly
connected subsets (this view is good that way).
4. Select smaller subsets (zoom in on details).
5. Interrogate specific nodes for their details.
6. Manually aggregate what my visual judgement suggests are
"clusters", building a hypergraph.
And all this without really knowing what the data is and what they are
really trying to show here. The more I look at it, the more I get out
of it (and the more questions I have). Does anyone else have this
experience? Or is everyone else equally puzzled by this kind of "map"?
- Steve
PS. Yes, Doug, I am avoiding a deadline, why else would I dive in so
deep on this!
------------------------------------------------------------------------
============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org