Modeling a User's Domain Knowledge
In an AHS the domain model consists of a set of concepts.
The user learns about concepts by reading about them and/or by taking
tests or performing assignments.
There can be a one-to-one correspondence between concepts and nodes
(a "fine grained" approach), but a single concept may also correspond
to a large set of nodes (a "course grained" approach).
Different techniques (value sets) are used to model a user's
knowledge about a concept:
- In the Boolean model a concept is either
known
or not known
by the user. This approach is only usable with
a fine-grained user model: each time a user reads a node, the concept
that corresponds to that node becomes known
.
AHA [DC98]
(used for this course) uses the Boolean model.
- In the discrete model a few knowledge-values are
distinguished. They can mean, for instance,
not known
,
learned
, well learned
, well known
.
A concept can become learned
by reading about it at a time
which may not have been appropriate, well learned
by
reading about it when the system decided it was the right time to learn
about it, or well known
by successfully completing a test
about the concept.
ELM-ART [BSW96a] and
Interbook [BSW96b] use such a discrete model
for instance.
- In the continuous model a very large set of possible
knowledge-values is used, for instance the interval [0..1], but some
systems [PDS98] use (whole) percentages as an
approximation of a continuous interval. The continuous model is most
useful in a course-grained knowledge representation. Each page about
a concept that is read contributes towards the knowledge about that concept.
Reading most or all pages and possibly also taking a test is needed
to reach the highest possible knowledge-value for a concept.
Systems that use a continuous model can also (potentially) make use of
the reading time in order to estimate how well a node has been read,
and of the fact that humans tend to forget over time.
Apart from knowledge per se most systems also log some navigation history
information as well. So about each concept not only a knowledge-value
may be known but also some information as to how the user obtained that
knowledge. (For each node the access times may be stored for instance.)
In general one may assume that for each concept an AHS stores a number of
attribute values, of which the knowledge-value is only one.