Databases Reference
In-Depth Information
do not have explicit semantic web links, the end-user still has to be fairly careful about what links
are chosen in order to have useful annotations.
over showing the possibilities for embedded links
books” . or music albums”
among menu athorn
z
Links
ges
”
all of your exisring
Web page
vant, informative and action
Semantic page
ta (additional links) behind l
kedness.”
Confessions on a Dar
Where there is no annotation at all, text-mining techniques can be used to find suitable points
that can link the text to semantic resources. The earliest example of this is in the pre-web Hypertext
system Microcosm (
Hall et al.
,
1996
). Most Hypertext systems, at that point (and HTML web
pages today), relied on the author of the source page to add some sort of link to the target material.
However, this means that the hypertext linking is limited to the author's knowledge of the potential
material at the point of creation; if the author does not know everything that could be linked to, or
if the material changes, then the authored material is incomplete or out of date. Microcosm instead
used automatic
external linkage
: the page author did not worry about what was going to be linked to
but instead, simply chose key words and terms for the document; then when any other document used
those terms they became live links. For example, imagine a variant of Wikipedia, the authors never
create links, but every mention of “Edgar Codd” in the text is automatically converted into a link to
could be used by the server to create links to more structured material, for example, in a University
system, course codes in meeting minutes could be linked to the relevant course documentation, or
in a commercial setting, product codes linked to the product information or ordering system.
This same principle of external linkage can be found in a class of systems called
data de-
tectors
that were originally developed in the late 1990s, including the Intel selection recognition
agent (
Pandit and Kalbag
,
1997
), Apple Data Detectors (
Nardietal.
,
1998
), Georgia Tech's Cy-
berDesk (
Wood et al.
,
1997
), and aQtive onCue (
Dix et al.
,
2000
). Data detectors work by examin-
ing some aspect of the user current context and then suggesting possible actions based on that. For
example, most current email applications recognise URLs embedded in the email message and turn
these into live links, and they are a form of primitive data detector. Data detectors also recognise
other features such as dates, names or postal codes and use these to create links to desktop or web
resources. For example, if a name is detected the user might be directed to a web service to lookup
