Information Technology Reference
In-Depth Information
to satisfy the preferences (wishes) of the user. The agent can inform the user of its actions
through the educational portal.
The difficulties, associated with the management of the pro-activity of our architecture,
result from the fact that the portal is designed for reaction of the user's requests. Therefore
the pro-activity can be managed only asynchronously and for this purpose we provide
development of a specialized service, which is to check a "mailbox'' periodically for
incoming messages from AV.
According our architecture, the reactivity and the pro-activity are possible if the
environment of the agents (Agent Village) remains not more passive. In order to be
identified, the agents need a wrapper (the environment), which "masks'' it as a web service
for the portal. In such a way the portal send the request to this service (masked
environment), which in its turn transform the request into an ACL message, understandable
for the agents. In a similar manner the active environment transform ACL messages into
SOAP responses, which can be process from the portal services.
The next assistants are developing in the first version of the AV node:
Evaluator Assistant (EA);
FraudDetector;
Statistician;
Intelbos
The Evaluator Assistant (EA) provides expert assistance to the teacher in assessment of the
electronic tests. In the Exam Engine a service is built for automated assessment of "choice
like'' questions. In the standard version of the architecture questions of the "free text'' type
are assessed by the teacher and the ratings are entered manually in the service to prepare
the final assessment of the test. In the cluster the Exam Engine calls the assistant (an
intelligent agent), which makes an "external'' assessment of the "free text'' type questions. In
the surrounding environment of the EA, the received SOAP Request messages are
transformed into ACL messages, understandable for the agent. Some of the basic parameters
of the messages are:
Text, which is an answer of a "free text'' type question.
Parameters for the used estimation method.
Maximum number of points for this answer.
The EA plans the processing of the request. In the current version of the assistant two
methods are available for estimation:
Word Matching (WM)
method - counts "exact hits'' of the keywords in the answer. The
minimum threshold of percentage match (i.e. a keyword to be considered as "guessed''),
which is laid in the experiments, is between 70% and 80%. Intentionally, the method
does not look for 100% match, in order to give a chance to words with some minor
typos also to be recognized. To calculate the points, offered by this method, a coefficient
is formed in the following way: the number of hits is divided by the number of
keywords. The actual number of points for the answer is calculated as the maximum
number of points is multiplied by this coefficient;
