Hi Georgi,
First let me underline that the following is not a detached theory, it
is very practical:
The web of data can support the clinician in his cycle of decision:
(a) The clinician makes measurements (in the broadest sense, also
speaking with the patient and looking at a picture is a measurement).
(b) The clinician focuses on those measurement results which are
interesting for his therapeutic decisions (feature extraction).
(c) The clinician compares these measurement results with experience.
At this he may use rules or models which are derived from common experience.
(d) The clinician decides for therapy, and measures the effect of his
decision, i.e. the cycle starts again with (a).
Good and large experience is very important for step (c).
The cycle of decision (measurements - feature extraction - comparison
with experience - decision) is also effective outside medicine: Before
every conscious decision we *compare* decision relevant data with
experience (or a model which is derived from common experience).
Experience says, at *similar* situations possibility X yields better
results than other possibilities, so we decide for possibility X. Even
if we try to decide best, our decisions are suboptimal due to limited
experience.
The web of data can be designed in a way, that it collects experiences
(also decision relevant measurements of machines) in a precise and
*comparable* way (much more precise and better comparable than text). So
the web of data can summarize experiences in well defined comparable way
for decision support.
For this a clear similarity relation is necessary. The natural way to do
this is a vectorial description of resources, i.e. quantification of the
resource's properties and regarding the result (a sequence of numbers)
as vector. After defining an appropriate metric (distance function) we
can calculate similarity of vectors by calculating the distance between
them - the less the distance, the more similar are the vectors and (in
case of good quantification) the original resources. Using HTTP URIs
allows that all domain name owners can define these vectors and
optimized distance functions.
Therefore i suggest to introduce standardized "Vectorial Resource
Descriptors" (VRDs) on the WEB - and it seems the best possibility to
integrate these in Linked Data. The paper
http://www.orthuber.com/wp1.pdf describes details. It is not completely
up to date, and though the basal content of the VRDs (and Vector Space
Descriptors - VSDs) is clear, I have not been sure about the syntax of
the RDF examples (Chapters 2.2.2 and 2.2.3 currently) - and I would like
to adapt the syntax to suggestions from the community.
So comments and suggestions are very welcome!
Best
Wolfgang
Georgi Kobilarov schrieb:
Yesterday issued a challenge on my blog for ideas for concrete linked open
data applications. Because talking about concrete apps helps shaping the
roadmap for the technical questions for the linked data community ahead. The
real questions, not the theoretical ones...
Richard MacManus of ReadWriteWeb picked up the challenge:
http://www.readwriteweb.com/archives/web_of_data_what_would_you_build.php
Let's be creative about stuff we'd build with the web of data. Assume the
Linked Data Web would be there already, what would build?
Cheers,
Georgi
--
Georgi Kobilarov
Uberblic Labs Berlin
http://blog.georgikobilarov.com