Probabilistic Relational Learner Models Based on Competence Maps

We present a proposal for a computational representation of competence maps that emphasises relationships of inclusion/part-of and specialization/generalization, and a generic approach to the construction of probabilistic relational learner models based on those competence maps, in which conditional probability tables are built on the basis of the kind of relationships between competences and, for the case of inclusion/part-of relationships, on the number of those relationships. We justify the use of noisy-or as a substitute for composite conditional tables produced by a competence being part of many other competences. Preliminary testing of both frameworks, for computational representation of competence maps and the construction of probabilistic graphical models from them, suggest coherence with reality.