Abstract:
On the basis of the model proposed by Rush and Wallace (1997) for elicitation of knowledge
from multiple experts, I seek to implement the model and test the accuracy of the generated
central network. The Multiple Expert using Influence Diagrams (MEID) is a technique for
generation of an aggregate knowledge representation, called the central network, from several
experts each representing ones knowledge in the form of an influence diagram. The main
advantage of this technique is that it does not rely upon group interactions.
The measures of the aggregate knowledge representation are the mean central network and the
dispersion coefficient of the expert influence diagrams from the mean central network. The
accuracy of the aggregate knowledge representation is measured by the confidence interval of
distribution of the distance between the central network of the real experts and the central
network of the bootstrapped samples of expert influence diagrams. In this thesis work, the goal
is to test the adequacy and accuracy of the generated central network of real experts.
The results of the tests show that the two parameters used to measure the aggregate knowledge
representation are not sufficient as a measure. There is the need of another factor, which is the
total number of vertices used by the experts. The number of vertices affects the confidence
interval because the central network of the sample networks depends upon the vertices that are
used by the experts. Using the total number of vertices used and the dispersion coefficient
from the central network, we can judge the accuracy of the aggregate knowledge
representation.